Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements; and to You under the Apache License, Version 2.0. 242273 PARTFIX Please support 256 Colors in Terminal xterm-16color Term.setEmulation() will now accept "xterm-16color" in additin to "xterm". Regardless of which one is set Term.getEmulation() will now return "xterm-16color" as it is the superset. To achieve this ... - Moved palette base numbers (PAL_*) from Term to Attr. The fact that the stored values are off-by-one is dealt with in Attr.foreggroundColor/backgroundColor() which simplifies Term.csetBG() and csetFG(). - Attr.BGCOLOr and FGCOLOR fields were widened from 5 to 9 bits in anticipation of 256 colors. - Instead of relying on case-based value adjustments in Attr.setAttribute()/unsetAttribute() to convert rendition codes, what clients pass xterm, to attribute values use a 'map'. Attr.rendition_to_pindex() uses it to map a rendition # to a palette index and is used from Term.rendition_to_color(). - Adjust Attr.setAttribute()/unsetAttribute() to use the 'map'. However ... Terminal in NetBeans will start with TERM=xterm regardless of the value of Term.getEmulation(). The immediate reason for that is code in NativeExecutionServicve.runTerm(). Archaeology will "blame" this: 274671:effd016fde86 - ilia Jun 9, 2014 fixed Bug #244954 - support xterm-color in Terminal This fix passes on the value of $TERM from the shell tha NB was started from on to NB Term. It doesn't work very well. On my system if I start NB from a shell $TERm is "inherited" to e.g. "xterm-256color", but if I start NB from the panel launcher $TERM is "xterm". Read on to see whay this happens. In fact the hard-coding of $TERM and ignoring of Term.getEmulation() predates that. What is going on? How $TERM is used ----------------- At first blush you'd think that a "terminal emulator", e.g. xterm, will set $TERM to something, e.g. "xterm", "xterm-16color, "xterm-88color" or "xterm-256color", according to it's abilities. Then, an application running under that terminal emulator will pass on the value of $TERM to libcurses which, in turn, will consult one of two kinds of terminal information databases, "termcap" or "terminfo" and use the value of $TERM to map libcurses semantic actions to actual terminal sequences. How this fails -------------- Suppose you run a very new xterm which supports "TERM=xterm-256color" on a very old OS the termcap/terminfo for which supports only "xterm". Then applications which feed "TERM=xterm-256color" to curses will get an error message. I expect that for this very reason 'xterm' does, no matter what, set $TERM=xterm. So how do applications take advantage of newer xterm features? How Linux solves this --------------------- On my FC21 'xterm' sets $TERM to "xterm" (you need to verify this by inspecting xterm src code or using strace, _not_ by echoing $TERM in a shell) but a whole hierarchy of _shell_ initialization scripts, arranges to set "TERM=xterm-256color". This works because the FC21 distro build knows that it has built xterm with 256color capability so it feels justified in overriding $TERM in it's shell initialization scripts. How this doesn't work --------------------- Supose you log in from a new system where $TERM is set to "xterm-256color" to an old system the termcap/terminfo of which knows nothing about "xterm-256color". Any curses app run on the remote system will complain about not knowing about "xterm-256color". What to do? ----------- The fix 274671:effd016fde86 - ilia Jun 9, 2014 fixed Bug #244954 - support xterm-color in Terminal allows NB users some leeway to set $TERM before starting netbeans and for now I'm not going to perturb that. The best solution I can think of is to have NB carry the appropriate terminfo file with it and install it locally or remotely and use $TERMINFO or $TERMINFO_DIRS to get applications to access the correct terminfo file for a given implementation of NB Term. Some applications, notably 'vim', don't seem to use curses but they _do_ honor $TERMINFO and $TERMINFO_DIRS. For a more in depth exposition consult private communication subjected "On the correct settingof $TERM". nobugid PARTFIX Enhanced info for AIOOB Exception in setCharacterAttribute() Hard to track bug. Catch the AIOOB thrown in setCharacterAttribute() and print out some detailed info to help track the cause. 242273 PARTFIX Please support 256 Colors in Terminal Factoring of Term.backgroundColor() and foregroundColor() in preparation for moving the decoding of color attribute into Attr. 242273 PARTFIX Please support 256 Colors in Terminal Renaming of variables in backgroundColor() and foregroundColor() in preparation for factoring. 242273 PARTFIX Please support 256 Colors in Terminal Switch to a single palette implementation. Previously Term used two "palettes" to map color indices into actual Color's. The two palettes were 'standard_color' and 'custom_color'. The former was strictly private and the latter was configurable via Term.setCustomColor(). The actual mapping was performed in Term.foregroundColor() and backgroundColor(). We now use a single 'palette' with domain bases specified by PAL_* constants: OLD NEW FG BG --------------------------------------------------------------------- standard_color[n] palette[PAL_ANSI+n] 30-37 40-47 custom_color[n] 50-57 60-67 OLD palette[PAL_BRIGHT+n] 90-97 100-107 NEW default foreground palette[PAL_FG] default background palette[PAL_BG] palette[PAL_RGB] palette[PAL_GREY] In short, custom colors now map to "bright" colors. These custom colors were part of a proposal to enhance DtTerm which never made it. Being non-standard mucking with them will have little impact except for how IO operations used it (See below). While the palette is populated, in initializePalette(), with RGB and GREY Colors no Interp handles RGB/GREY sequences yet. The encoding of colors in Attr cells is still the old way (which is offset by 1 to allow encoding of "default" as 0) and the adjustments are all done in Term.foregroundColor() and backgroundColor(). This will be optimized later. Therefore, setCustomColor() has been deprecated. Overriding Component -------------------- Now that we have a single palette, setting of FG and BG colors should keep it up-to-date. That is done by overriding Component's setForeground() and setBackground(). This has to be done with due mindfluness of 'reverse_video'. Reverse video tricks -------------------- When reverse video is requested we swap palette[PAL_FG] and palette[PAL_BG]. This means that when we handle setForeground() or setBackground() we need to be mindful of the value of 'reverse_video'. Inherit from L&F ---------------- The default fg and bg colors are now initialized using UIManager.getColor("TextArea.foreground"); UIManager.getColor("TextArea.background"); Term.setCustomColor() never worked ---------------------------------- ... because I had these InterpProtoANSI.dispatchAttr() functions which screened the actual attribute values and none of them accepted attribute values in the range 50-57 or 60-67. This must've worked at some point in the past. Then, when I was working on proper xterm emulation, I introduced dispatchAttr() and neglected the "custom" atribute values. The only serious place which used Term.setCustomColor() was the mechanisms in TerminalInputOutput which supported IO-style setting of Colors. See, for example, TerminalInputOutput.customColor() which added 50 to the color code. This has now been switched to add 90. To see it "not work" run the project lib/terminalemulator/examples/TermExample Then choose Terminals->TestTerminalWithRichNativeExecution Press OK. At the very begining You'll see GREETINGS green Choose blue Unchoose red Select chosen blue # SHOULDA been Color.ORANGE The use of setCustomColor() in this fashion is temporary. Eventually TerminalInputOutput should be able to use direct RGB color setting using sequences like "ESC[48;2;...m". Bug in Term.backgroundColor() ----------------------------- In the !reverse case we never handled the case of bcx = 0. Incidental ---------- - Get XTermTestSubject to force fg/bg colors on the xterm overriding XTerm properties so that it matches other TextSubjects. - Enhance Test_attr to test "bright" attributes as well as running in "reverse video" mode. 242273 PARTFIX Please support 256 Colors in Terminal Redid Attr class into an enum as I will be playing more with these bitfields in order to accomodate more colors. Original Attr class saved as AttrSave. The enum approach is a teeny bit slower since field params aren't hard-coded in the instruction stream as constants and one has to do instance field access but I did benchmarking as well as profiling and Attr class didn't and doesn't contribute much to hotspots. I.e. 0% in the profile. "test bc" was registered as "attr" by mistake so when I ran test "attr" I got test "bc". Fixed so "attr" runs "attr". 242273 PARTFIX Please support 256 Colors in Terminal Fix benchmarking/statistics gathering escape sequences. They were implemented in InterpANSI (ACT_PRINT) but with the switch to xterm a while ago they should've been moved to InterpProtoANSI. Fix Term.indent(). Doing println("\t") for indentation makes no sense. It should've been print(). nobugid - Tinker with TermApp's runargs Had added VM options to set AA fonts to help figure out complaints from Tim. Looks like the VMOptions textarea doesn't like the options separated per line and when I run it the VM complains. But it worked once! ================================================================================ nobugid - Adjust spec version to 1.37.6 One of the things that triggered an apichanges was the introducton of TermOptions.PROP_ALT_SENDS_ESCAPE. It turns out none of PROP's in TermOptions is really used externally. I.e there's no API which sets of gets them and there's not property notifications that take them as parameters them so might as well make them all private. ================================================================================ nobugid - Massive tip cleanup 242439 - Terminal -- Clear preserves position of the prompt Fix is easy. Just have the Clear action call Term.clearHistory() instead of clear(). User Visible Changes -------------------- Clear action will clear history, selection and move cursor to 0,0. Added a Ctl-Shift-L accelerator for Clear. It can't be Ctl-L like it is in the regular output window because Ctrl-L might mean something to the application running in the terminal. Allother terminal accelerators are Ctl-Shift for the same reason. 236268 - terminal ignores Alt-f, Alt-b keys Introduced new Term property, altSendsEscape. It is matched with a TermOptions property of the same name and a check box in org.netbeans.modules.terminal.nb.TermOptionsPanel. I was surprised to find out that TermOptionsPanel under org.netbeans.lib.terminalemulator.support was cloned, for reasonably good reason, into terminal.nb/TerminalImplementation. So for the moment, the original TermOptionsPanel doesn't support altSendsEscape yet. Syncing the two implementations will be a future project. Term's altSendsEscape property affects the behaviour of Term's screen's charTyped() KeyListener. It is based on xterm behaviour. You can read more about it in the javadoc for Term.setAltSendsEscape(). User Visible Changes -------------------- If altSendsEscape is checked (the default), when Alt is used as a modifier for some other key, say K, it will be converted to an ESC followed by K. If altSendsEscape is not checked, when Alt is used as a modifier, characters in the range 0-127 will be "shifted" to 128-255 by adding 128. This allows the entering of "8bit ascii", "accented", characters used predominantely in the "latin" charsets. It takes a bit of care to actually see this work. For example, my personal locale has LANG=en_US.utf8 but everything else, LC_*, is "C". Java looks at $LC_ALL to decide it's Charset and for "C" it chooses US-ASCII which is a 7-bit ASCII encoding so anything modified by Alt showed up as '?' for me. The remedy was to set $LC_ALL to en_US.utf8. There are also some shananigans that 'bash' pulls such that if you type Alt-e at bash you won't get the accented 'e' but if you use 'cat' or 'od' or 'vi' things get echoed properly. The Mac ------- Apparently my earlier assumption that "altIsNotMeta false" doesn't hold for the Mac for it has a dedicated Meta key, the Apple key, distinct from Alt. However, I don't have a Mac so it'll take me a bit to figure what's the right thing to do. Suggestions welcome. Terminal options dialog ----------------------- The usual accelerator choosing game. I freed S by using z for FontSize and assigned it to AltSendsESC. ================================================================================ nobugid - Track gnome-terminal changes in TermTester nobugid - Move handling of [t (ACT_GLYPH) from InterpANSI to InterpDtTerm. nobugid - Make background of images in the glyph gutter be transparent instead of white + Misc. \n fixes. nobugid - TermTester updates - Switch to javac.{source,target}=1.8 - Look for xterm and gnome-terminal in /bin instead of /usr/bin. - New test commands 'mark' and 'glyph' to test Term's ability to place glyphs in the glyph gutter. - Enhance 'tab' test to help debug problems with HT handling. nobugid - Allow ActiveTerm to react to right mouse button. ... so we can do context menus for active regions. ================================================================================ nobugid - Track various NB API changes These are in projects in the examples directory. - JNA Structure now needs getFieldOrder() implemented. - TermListener now needs cwdChanged() implemented. nobugid - TermTester These are testing utilities and various specs that I had kept separate. It's high time they got integrated. See http://wiki.netbeans.org/TerminalTestingAndTroubleshooting for an introduction to the use of these utilities. nobugid - termcap and terminfo sequences ... for xterm, ansi and dtterm added to doc-files/ 238225 - Midnight Commander and re-size breaks terminal Looking at traces from Term MC sets the margins (op_margin()) but never resets them (reset/default margins always track the window size). MC does issue the sequence \ESC[?1049l which is really the combination of ... 1047 Revert to normal screen buffer, clearing screen if switching away from alternate screen buffer. 1048 (DECRC) Restore cursor. But AFAICT margins are not a property of a "screen" nor of a "cursor" ("cursor" is actually a collection of attributes) so neither of the above play a role with margins. Instead, it seems, margins need to be reset on resizes. I checked with 'terminator' and 'konsole' and they reset the margins on resizes only. This is easy to fix by adding a private Term.resetMargins() and calling it wherever st.rows gets modified. BTW Term doesn't support alternate screens yet. This might explain some of the other oddities when exiting MC, vi, man etc. nobugid - Track API changes: TermListener.titleChanged() This is in examples/ projects which got missed when titleChanged() was added. TermApp now shows it's title. nobugid - Default $TERM for TermApp -> xterm nobugid - PrintStatsAction for TermApp This is all to help debug bug 238225. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset : 265081:ccb1fab67b20 summary : Switch from "ansi" to "xterm" as the default terminal emulation type ($TERM) 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals op_ind() and op_cud() are now part of Ops. op_line_feed() now implemented in terms of op_ind(). Fine tune op_cuu(0 and op_cud(). op_full-reset() clears history. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset : 242713:1841c687fb7b Refactoring before tackling IND and CUD. Implement IND (Index) \ESCD scrolls line feed \LF == IND CUD (CUrsor Down) \ESC[%dB doesn't scroll using op_ind() and op_cud(). op_line_feed() maps to op_ind() for bwd compatibility. This fixes a bug where \ESC[%dB would previously scroll. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset : 242712:a9901f0515a4 Implement RI (Reverse Index) \ESCM scrolls CUU (Cursor up) \ESC[%dA doesn't scroll using op_ri() and op_cuu(). op_up() maps to op_ri() for bwd compatibility. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals Implement CBT (Cursor Backward Tabulation)\ESC[%dZ op_cbt CHT (Cursor Horizontal Tab) \ESC[%dI op_cht 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset : 242710:7d3b4c76fb2a Implement ED (Erase in Display) and reimplement \ESC[%dJ in terms of op_ed() instead of op_cd() or op_cl(). 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset: 242709:a0c447ec4b2c Fix EL (Erase in Line) and ECH (Erase CHaracters) especially wrt retaining background color in erased regions. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset : 242708:8e56a4a57a11 While fixing cursor show/hide discovered that the real problem is that for DEC private actions didn't correct handle multiple ;-separated numbers - last one was dropped. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset : 242707:b9d72096f34f Implement \ESC[%dG CHA \ESC[%dX ECH \ESC[%dd VPA 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset : 242706:f2e980336ed7 - Handle unsupported Toolkit.getLockingKeyState(). See comment in InterpProtoANSIX.numLock(). - Accept multiple numbers for \ESC[? - Parse \ESC[> family in InterpXTerm but leave actions for later except for \ESC[>c. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset: 242705:ba366f6407ec Emit correct sequences for function, edit, numpad and arrow keys for InterpProtoANSIX ("xterm" and "dtterm"). This comes at a slight price of regresions in Term accessibility - for the above terminal types only. PageUp, PageDown, Ctl-UpArrow and Ctl-DnArrow will no longer scroll according to Swing L&F conventions. This is acceptable if you consider Term to be the ultimate keyboard accessibility mechanism where terminal conventions override those of Swing L&F. Implement DECPAM, DECPNM and DECCKM to control alternative key sequences that get sent from above. InterpANSI also emits special sequences for arrow keys and Insert and Home keys. InterpProtoANSIX now keep's it's own state to keep tyrack of PAM and CKM as opposed to using State. Implementation depends on new method Interp.keyPressed(KeyEvent e) which, in turn, depends on Ops.send_chars(String sequence) Because state is now kept in interpreters had to add Interp.softReset(). 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset: 242704:6f3266eb5908 - \ESC[%dl implementation moved from 'ansi' to 'protoansi'. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset : 242703:674b8a6f08a4 - \ESC7 and \ESC8 implementation moved from 'ansi' to 'protoansi'. - Implement all codes (012) for \ESC[%dK. Uses new Ops.op_el(int code). Ope.op_ce() now just delegates to op_el(0). 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals - Implement sequences for ProtoANSI \ESCn \ESCo \ESC(B \ESC(0 \ESC)B \ESC)0 \ESC*B \ESC*0 \ESC+B \ESC+0 These utilize Ops.setG() and selectGL(). - Font selection is now done in a slightly more complex, two-step, process. Sequences \ESC(, ), * and + assign a font to one of graphics sets G0, G1, G2 or G3. Then the sequences \SI, \SO, \ESCn and \ESCo choose one of G[0123] as the rendered font. All this info is now kept in State and State.font() returns the "current font". - For "ansi" emulation fonts are set using \ESC%dm. Previosuly this used to go through Ops.set_attr() but now it goes via Ops.op_setG() in InterpANSI.dispatchAttr(). - Attr used to store font info but that was only used by State not by actual buffer cells. Now that font state is more complex it's stored explicitly in State and all Attr font handling becomes dead code. Removed the dead code. - Fixed rendition of the diamond ACS graphic character. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals - Increment spec version to 1.25 - InterpProtoANSI and InterpProtoANSIX were supposed to be package private. - Add ability to customize Alternative Character Set (ACS) encoding by Interp. The key is char Interp.mapACS(char) which will take an interp-specific encoding and return a canonical encoding or '\0' which means the passed in char does not encode an ACS. The canonical encodings correspond to curses ACS_ variables and characters used by the 'acsc' terminfo attribute. For example "infocmp ansi" will yield: acsc=+\020\,\021-\030.^Y0\333`\004a\261f\370g\361 ... and therefore mapACS for ANSI should return '-' when passed \030. This is the main fix needed for 187345. That is, if "xterm" emulation is chosen for a Term it will handle the rendition of graphic characters correctly. However handling of _switching_ to the ACS font will take a bit more doing because "ansi" terminals honor \ESC10m and \ESC11m but "xterm"s require \ESC(0 and \ESC(B and then \SO and \SI. - Add ability to customize responses to attribute codes by Interp. The sequence \ESC%dm is a general attribute setting mechanism. However, the values accepted by "xterm", "ansi" and "dtterm" differ wildly. Yet, fortunately, there's no overlap in functionality so we can still leave the ultimate implementation of the code up to Term/Ops/Attr. Handling of the actual value of %d is now delegated to Interps via boolean InterpProtoANSI.dispatchAttr(AbstractInterp ai, int n) which check for valid values of 'n' and call ops.op_attr(). Incidentally added a couple of missing codes to Attr.setAttribute(). 3 and 6 are now accepted although they fall back on simulations. This is all private to the terminalemulator package and only applies to Interp's derived from InterpProtoANSI. - Using an intermediate private Term.mapACS() as a trampoline between mapChar() and Interp.mapACS() didn't quite work out right. Need to test for '\0' directly in mapChar(). - Added additional ACS support for arrows and blocks. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals Distribute "text parameter" sequence interpretation properly between InterpXTerm and InterpProtoANSIX. 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals Barebones xterm support changeset: 240886:95ea4996d8d5 Introduced package scoped class InterpXTerm. Term.setEmulation() now accepts "xterm". xterm sequences are similar to those of ansi and dtterm so this barebones implementation isn't really differentiated. In particular xterm emulation doesn't handle alternative renditions, the main issue in this bug, correctly yet which is why ansi should still be used as the default emulation. TermApp enhanced with a -t option which allows setting of the terminal type to one of dumb|ansi|dtterm|xterm. nobugid - Cleanup and better naming for examples directory changeset: 240885:eaea739e914d - demosrc/lib.termsupport/nbterm -> examples/TermApp - demosrc/lib.termsupport -> - demosrc/Suite -> examples/TermSuite - demosrc/Examples -> examples/TermExample - demosrc -> examples nobugid - sequence logging for debugging and testing changeset: 240884:6efe424d9990 API enhanced with the following: void Term.setSequenceLogging(boolean) boolean Term.isSequenceLogging() Set Term.getCompletedSequences(); Set Term.getUnrecognizedSequences(); void Ops.logCompletedSequence(String); void Ops.logUnrecognizedSequence(String); The idea is to turn this logging on, run some terminal-heavy apps like vim, rogue/nethack, mc, alsamixer etc and see what escape sequences were actually processed and which were, silently, ignored. This can help with determining coverage of a given application. A corresponding change in the Terminal module enables logging if -J-DTerm.debug=sequences is used on the NB cmdline. The property will additionally enable a Terminal context menu item "Dump Sequences" which will dump the above respectively into /tmp/term-sequences-completed /tmp/term-sequences-unrecognized So, if a user complains about some fancy sequence not working, the thing to do is to run NB with -J-DTerm.debug=sequences, run the misbehaving application and then issue "Dump Sequences" from the terminal context menu. "term-sequences-unrecognized" is likely to contain a sequence that is not yet implemented. Implementation: protected String InterDumb.ctl_sequence was replaced with a more comprehensive and efficient private StringBuilder ctlSequence 207965 - Terminal incorrectly displays prompt string 187345 - wrong characters in Terminals changeset : 240883:4e7a19ec8f00 Preparation for introduction of xterm emulation. Introduced a table to track control sequences common to ansi, dtterm and xterm: lib.terminalemulator/doc-files/sequences Introduced InterpProtoANSI and InterpProtoANSIX. - InterpProtoANSI is intended to factor Interp states and actions that are common to ansi, dtterm and xterm. - InterpProtoANSIX is intended to factor Interp states and actions that are common to dtterm and xterm. nobugid - debugging printfs for Failed and Successful sequences changeset : 240882:a7abd9388502 The idea is to log seen control sequences, both those which succeeded and those which failed. Later this output will be controllable using Term debug flags. nobugid - Get lib.richexecution and TermApp projects compilable again changeset : 240881:d942f87bcfdd Fixing of various dependencies and project file regenerations. nobugid - move documentation out of src directory changeset: 240878:8323630292f9 ... and into lib.terminalemulator/doc-files ................................................................................ nobugid - fix op_ce (clear to end of line) so there are no "boxes" when running nethack. This in response to Jesse's comment that nethack doesn't work under term. See bug #192779 Comment #1. After fixing this, and maybe thebelow bugs, I haven't had problems running nethack on linux. nobugid - ensure ${cluster} is initialized before clean nobugid - Term debugging switch from UI The "Start Terminal" dialog which pops up on Terminals-> TestTerminalwithRich/NativeExecution now has a debug checkbox. Entries in the ControlWindow now have a "debug: on off" control. These allow enabling and disabling of Term debug flags. nobugid - don't NPE when creating a Terminal and ControlWindow is closed. 187345 - wrong characters in Terminals Partial fix ... A program renders alternative characters as follows. It first needs to select the desired font via "ESC [ m". Where ranges from 10 to 19, 10 meaning "default" font. About the only interesting font in widespread use is 11, the so-called DEC graphics character set. See the right two columns in http://vt100.net/docs/vt220-rm/table2-4.html Then the program needs to emit the right character codes. These vary from terminal type to terminal type and are described by the 'acsc' entry in the terminfo DB: cd /usr/share/terminfo/a infocmp ansi The mapping described by 'acsc' is for _output_. E.g. to render the DEC graphic character '~', i.e. ACS_BULLET, the program has to emit 0304 for an ANSI terminals and '~' for xterm. To that end ... ... introduced support for alternative font character attributes. These are set by codes 10-19 of the ANSI "ESC [ ... m" escape sequence, 10 being the default font. These codes are normalized to a 0-9 range and stored in a new 4-bit-wide FONT field in Attr using Attr.setAttribute() and setFont() and subsequently accessed via Attr.font(). How are various fonts rendered? We work on the assumption that unicode will contain a glyph for any desired graphic. This reduces the problem to mapping a character coming into op_char() to the appropriate unicode "glyph". This is done by Term.mapChar(). mapChar() only handles additional font attribute 1 (aka ANSI character attribute code 11). It does so in two steps. First it maps the incoming character to the Alternative Character Set (ACS) based on http://vt100.net/docs/vt220-rm/table2-4.html These are canonical VT-100 characters. It then maps the canonical VT-100 character to an explicit unicode character representing the desired graphic. Still to do ... - Term.mapACS() needs to be an abstract method of Interp. - Not all VT-100 graphic characters have high fidelity representations in unicode. A more sophisticated system would create it's own scaled glyphs. And all of this hasn't solved the original problem of the fancy zsh prompt. I can get almost everything right except that there are too many spaces on the second line of the prompt causing the "date" portion to wrap. But I think this has to do with the terminfo entry for 'ansi' or a curses bug on my FC12. Will have to check on some other platforms after pushing. 188024 - single last character with different attributes not rendered An off-by-one error in the main loop termination test in Term.paint_line_new(). 187345 - wrong characters in Terminals Introduced a version of StreamTerm.connect() that takes a charset like "ISO-8859-1" or "UTF-8". This allows for overriding the default system encoding. Sort of analogous to Project source encodings. ================================================================================ May have missed some entries here ... ================================================================================ nobugid (Towards more InputOutput functionality) Introduced IOResizable and IOEmulation to support Term-related functionality. IOREsizable allows capturing terminal size changes and forwarding them to the relevant ioctl(). IOEmulation allows querying the actual terminal emulation provided and assigning it to $TERM. It also proviedes the disciplined property which describes whether the InputOutput provides it's own line discipline or relies on an external agent (usually a pty). TerminalInputOutput implements these new capabilities. TerminalIOProviderSupport, which is code common to many of the demo actions in demosrc, now uses the above InputOutput-based functionality instead of accessing a Term directly. ================================================================================ nobugid (Streams from StreamTerm) While experimenting with IOExecution it becaame clear that there would be some merit for StreamTerm to provide actual Streams the way InputOutput does. To that end StreamTerm provides: Reader getIn() Writer getOut(). This simplifies the implemenatation of TerminalInputOutput. Or rather, factors some of the implementation into TerminalInputOutput. nobugid (external hyperlink generation) Terminal can now recognize externally generated hyperlinks (per http://wiki.netbeans.org/TerminalEmulatorHyperlinking). You can generate such hyperlinks using lib.terminalemulator/demosrc/terminal/examples/make_filter.awk as follows: make | awk -f make_filter.awk In order to process such hyperlinks one has to call Terminal.setHyperlinkListener which takes a new interface HyperlinkListener. This is a bit cleaner than the older way of doing this as described in http://wiki.netbeans.org/TerminalEmulatorHyperlinking?version=6 CommandmermDirectAction (aka Command with pty and direct access to Terminals) sets up a hyperlink listener. nobugid (Experimentation with IOExecution) Have a temporary IOExecution per Tomas Holy's original proposal in terminal/ioprovider. Moved Program/Command and Shell into it's own package inside richexecution. Program used to be based on ProcessBuilder and even returned it. Now it's a pure data object and PtyExecutor builds the ProcessBuilder on demand. TerminalInputOutput now implements it. nobugid (more: track new I/O API's) terminal/example module has these: - IOFeaturesAction creates a Term base InputOutput to demonstrate various IOProvider features: - IOPosition - IOColorLines - IOColors - IOTab (this uses sunsky.png) They are basically implemented in TerminalInputOutput. nobugid (track new I/O API's) TerminalIOProviderSupport.getIOProvider() now uses IOProvider.get("Terminal") instead of it's own iteration through lookup. The old code is relegated to getIOProviderClassic(). terminaIOProvider implements getName() which returns "terminal" so IOProvider.get() can find it. TerminalProvider.createTerminal() variation which takes an IOContainer. ioprovider.TerminalInputOutput variation which takes an IOContainer and calls TerminalProvider.createTerminal(). Commented out stubs implementing IO "features" (IOShuttle is used there). Terminal can now be contained in a TerminalContainer as well as an IOContainer. It doesn't fully implement all IOContainer.CallBacks. In particular "close" semantics are still undefined. The following behave differently depending on where a Terminal is embedded in: select() setTitle() setActions() can only be done at construction time with IO. find() not supported for IO. closeWork() TerminalContainer.reaped() -> removeTerminal(). select, activated and deactivated passed thorugh from TerminalContainer to Terminal via callBacks. TermTopComponent overrides TC componentActivated()/Deactivated() to support this. nobugid (prep for API review) Started arch.xml for rich execution. Command, PtyExecutor made final. Platform, Util made pkg private. TermExecutor made final and now delegates to PtyExecutor instead of inheriting. nobugid (Font chooser -- pass1) Users have long chafed at the limited choice of fonts, basically "monospaced" in the output window both when it was based on Term and afterwards. See, for example, IZ's 29604, 40033, 43165, 45174 55455, 87536. The main reason for this limitation is a bit different for Term and output2. Term is a _terminal_ based on rows and columns and in principle only makes sense with fixed width fonts. Fixed width fonts dramatically pseed up layout and rendering of text. This is particularly important for output2 which has to deal with "unlimited" buffer sizes. However, we need not restrict ourselves to "monospaced". A typical system has a large palette of fonts and some of them are bound to be fixed width. This project adds a font chooser to TermOptionsPanel which allows the user to choose from among all the available fixed width fonts. However, we have a slight problem in that Swings fonts are not explicitly characterized by whether they are fixed width or not. So, we decide for ourselves by checking the widths of the first 256 characters and if they are all equal we consider that font to be fixed width. A more forceful approach would be to render variable width fonts in the fixed cells of a terminal. Presumably one can find the maximum width and use that as cell width. Rendering has to be done on a cell by cell basis and Term doesn't do that yet. The font chooser has a checkbox though for enabling non-fixed-width fonts in it's palette. To satisfy all of this ... - TermOptions now has a font property as opposed to just fontSize. - Term has a property, fixedFont, which governs what kind of font it will accept. If it's set to false term behaves as before just using the given fonts style and size and applying it to monospaced. If it's set to true Term will accept any font assuming that the user passed it a fixed width font. More works needs to be done: - One should be able to independently set the font size. - TermOptions' font property isn't properly saved and restored? - Whether variable width fonts are allowed should be part of TermOptions. ================================================================================ nobugid (Solaris work) - lib.richexecution - class CLibrary enhanced to work with Solaris. - Viable process_start-solaris-intel.zip created. - Viable process_start-solaris-sparc.zip created. - Verified that JNA code and the process_start's work under both 32-bit and 64-bit VM's. - Existing (zipped) process_start-linux-intel, whcih was built on FC6 crashes with a SIGFPE on newer linuxes like SLES10 or SuSe10 - lib.termsupport.nbterm - Introduced nbterm64 in order to test stuff under 64-bit VM's. - The distribution now contains all versions of process_starts which it gets from the zipfiles. ================================================================================ nobugid (Mac work) - lib.richexecution - Introduced mac-intel as a platform in build.xml which now builds process_start-mac-intel. - OS.platform() returns "mac-intel" to match. - A viable process_start-mac-intel.zip created. - In process_start.c need to explicitly assign a controlling terminal using TIOCSCTTY. We do this only if TIOCSCTTY is defined which should make the code platform neutral enough. - In JNAPty instead of getpt() or explicit opening of "/dev/ptmx" we use posix_openpt(). - New class Platform. OS.platform replaced with Platform.platform(). - Platform sensitive JNA LIbraries ... - PtyLibrary and ProcessLibrary interfaces merged into CLibrary class. CLibrary delegates to platform-specific Library's and initializes the constants according to Platform. - lib.termsupport - In TermExecutor.MyTermListener verified that setting TIOCSWINSZ via the master works on the MAc. - lib.terminalemulator - On the Mac VK_ESCAPE doesn't generate a keyTyped event() so need to simulate it using keyPressed(). Introduced charTyped() to factor keyTyped() and keyPressed() processing. Introduced boolean onMac(). ================================================================================ nobugid (Options support and UI) - lib.termsupport.TermOptions contains options information for a Term. - lib.termsupport.TermOptionsPanel is a generic JPanel for viewing the above. - org.netbeans.modules.terminal uses TermAdvancedOption to provide UI for options in NB under Tools->Options->Miscellaneous->Terminal. - TermApp (nbterm) has An Options context menu action and brings up an options dialog which is persisted in ~/.java/.userPrefs/nbterm. ================================================================================ nobugid (get working on Windows again) - richexecution - build.xml now recognizes windows (XP) as a platform - Use ${user.name} instead of ${env.USER}. - Command needs to pass /c instead of -c on Windows. - Pty constructor throws UnsupportedOperationException on Windows. - PtyExecutor.start() needs to initialize pid to -1 so that PtyProcess knows to not use unixy signals (it tests for -1). - On Windows PtyProcess should return the processes streams not the Pty's because there is no Pty. At some point we'll be able to do RAW Pty's on Windows and then this won't be neccessary. - termsupport - set the pty Mode to NONE on Windows. ================================================================================ nobugid (bugs in line insertion/deletion w and w/o margins) Symptom: vi a file in a 24x80 terminal. ^D line 23 isn't refreshed properly. Fix in Term.OpsImpl.op_al(), op_dl(). Seems to have fixed another symptom where fast ^D'ing when running 'vim' resulted in the "middle" line having lots of []'s. nobugid (towards distributing demo NBM) - Adjust Suite/build.xml to special-case the creation of lib.terminalemulators nbm's because it can't be added to the suite. - Add extra.module.files to RichExecution's project.properties to ensure that process_start* ends up in the NBM. - Enhance richexecution.PtyExecutor to "chmod u+x" process_start* because when they are extracted from an NBM zip cannot maintain their execution permission. - Module collateral information (Description home page etc.) filled in. Suite/build.xml richexcution/project.properties richexcution/PtyExecutor.java richexcution/Bundle.properties termsupport/project.properties termsupport/Bundle.properties terminal/example/project.properties terminal/example/Bundle.properties terminal/project.properties terminal/Bundle.properties terminalemulator/project.properties terminalemulator/Bundle.properties nobugid (i18n-check: misc. warnings in terminalemulator) terminalemulator/Term.java terminalemulator/Buffer.java Started a debug/test infrastructure although it's not in the mainline NB src code yet. The following were done to support it ... nobugid (Suppport for raw pty's on linux) On Solaris one makes a pty raw by just not pushing the stream modules ptem ldterm and ttcompat. On linux pty's are be default non-raw. One can make them act like a raw terminal by using cfmakeraw(). It only sets up a termios structure; one still needs to do a read/modify/write using tcgetattr() and tcsetattr(). So added all of these to richexecution.PtyLibrary: class Termios Linux only tcgetattr() tcsetattr() plus relevant constants cfmakeraw() JNAPty.assigFd() and getFd() moved to new Util class. JNAPty now uses the above functions to do a read/modiy/write for raw pty's. richexecution.PtyLibrary.java richexecution.JNAPty.java richexecution.Util.java nobugid (nbterm: couldn't find itself if executed through a soft link) Had to enhance the "find yourself" stuff at the beginning. nobugid (nbterm: xterm-like -e and -geometry flags) The terminal debugging/testing infrastructure fires up nbterm and a reference existing implementation like xterm or konsole and broadcasts sequences to all for visual comparison. Needed to add -e and -geometry to nbterm for this to work. -e used to mean "try error detection mode". That is now renamed to -E. nobugid (TermExecutor debugmode to emit more debugging stuff) termsupport/TermExecutor.java ================================================================================ 151644 (Return of the terminalemulator) Second large chunk of work. Additional functionality and a more conventional module and pkg organization. See http://wiki.netbeans.org/TerminalEmulator http://wiki.netbeans.org/TerminalEmulatorDemoModuleOrg ================================================================================ nobugid (Term and ActiveTerm support for hyperlinks) Term will now recognize the sequence ]10;;; as analogous to text and create hyperlinks. See http://wiki.netbeans.org/TerminalEmulatorHyperlinking for usage details. In addition to interpretation of the sequence in InterpDtTerm and it's implementation in Term.op_hyperlink ... This required a bit of support internally to help save and restore text attributes of regions enclosing hyperlinks: ActiveRegion.getParentAttrs() ActiveRegion.setParentAttrs() Term.attrSave() Term.attrRestore() nobugid (de-publicize RegionManager) 24760 (Eye candy for hyperlink navigation.) Partial fix. When the mouse hovers over a hyperlink the cursor shape changes to a pointing finger. ================================================================================ nobugid (TermSupport works on solaris now) - OS.UNIX -> OS.LINUX + OS.SOLARIS - JNAPty needs to do some ioctl(I_PUSH's) only on solaris. nobugid (TermSupport handles csh correctly) The call to setpgrp/setsid inside pty_bind pulled up into PtyProcess. See comment is wrappedCmd(). ================================================================================ nobugid (StreamTerm to handle IOExceptions better) See comments in StreamTerm.OutputMonitor.run(). ================================================================================ tag: ivan_25 nobugid (de-publicize all ACT_ classes) All Interps use internal Actor classes using the naming convention ACT_. These classes were protected _and_ final and an eyesore in the javadoc. Made them package private. 124612 files lost when terminalemulator was moved to cnd The transfer of termulator from core to cnd was incomplete in two ways: 1) Several files were dropped: ReleaseNotes.ivan.txt (this file) demosrc/buildtool/* demosrc/telnet/* test/unit/src/org/netbeans/lib/terminalemulator/TermTest.java These files have been restored and brought up-to-date. 2) Some changes were committed to CVS after the copy to cnd but before the core copy was deleted so they got "lost". These changes (tags ivan_24, ivan_23) have been reintroduced. Some more changes were never committed to CVS and missed both the core-to-cnd copy as well as the CVS-to-Hg transition. These are documented below. nobugid junit test failure on textWithin() In the process of reintroducing TermTest.java found that it fails. nobugid Abstracting of Buffer and Line classes In order to be able to alter the implementation of Buffer, for instance to have it use java.nio.Buffer's like output2, have to make it be more abstract. To that end ... - Enhanced Buffer.printStats() to provide more detailed statistics. - Line.glyph_glyph -> Line.glyphId - Line.glyph_rendition -> Line.backgroundColor - new property glyphId - new property backgroundColor - - Line.charArray + Line.accumulateInto - + Line.charAt(), charAtPut(), getChars() - Lot of code in Term.java used to pass a char buf[] around which used to be a pointer directly into a Lines storage array set in paint_line_new(). Now it uses Line.getChars() and a local xferBuf in myDrawChars(). nobugid Make Interp public On [fall 2007] Yarda, in order to satify some static code style checks, had, instead of making Term.setInterp(), getInterp() public, opted to make class Interp pkg private. Java apparently allows a sub-class of a package class to be passed to a parameter of the type of the package class. I.e. Interp is pkg private, MyInterp is public and extends Interp, ergo a public setInterp(Interp) is useful. Instead of quibbling, making Interp, setInterp() and getInterp() be public. ================================================================================ tag: ivan_24 These changes were driven by trying to get Midnight Commander (mc) to work under term. mc, with it's heavy dependence on terminal graphics, seems like a good litmus test. With these fixes we're doing fine on output, except for "graphical character" rendering. However, mc isn't very usable due to heavy dependence on function key and mouse event processing which I leave for another day. nobugid Handle "set text parameters" escape sequence ESC ] ; BEL Allows for various terminalemulator application text values like icon name, window title and current working directory to be set. We only handle these sequences so output from mc doesn't mess up the screen. While the sequences call new methods of class Ops, op_icon_name(), op_win_title() and op_cwd(), a terminalemulator _Application_ would still need some sort of listener mechanism to adequately handle these requests. Includes InterpDtTerm.ACT_DONE_COLLECT2. nobugid Factoring of InterpDtTerm.ACT_DEC_PRIVATE Mainly as I was exploring sequences having to do with enabling of mouse reporting. See http://www.xfree86.org/current/ctlseqs.html#Mouse%20Tracking ================================================================================ tag: ivan_23 nobugid Solitary attributed character in last column not rendered. scenario: Bring up 'vim', enter "aaa", enter ^V. A '^' will appear. In some environments the caret is blue and in such cases the caret isn't rendered. cause: In Term.paint_line_new() in the case where we use runs we bail out too soon because of this test: if (rend+1 >= lastcol) break; fix: Use > instead of >=. nobugid TAB inserts spaces when it should only move the cursor. This became clear in a curses example submitted by a customer. Where curses uses TABs as a quick way to move around. Fixed in Term.OpsImpl.op_tab() nobugid Handle ASCII SO and SI SO == ^N == Shift Out == LS1 == as switch to VT100 graphical characters SI == ^O == Shift In == LS0 == ae switch to default rendition These were not handled and were just echoed, throwing off curses screens. Term.OpsImpl.op_as/ae() handle these. They are now absorbed but there is no real support for graphical characters yet. Unicode supposedly has 32 codes for them U+FDD0 thru U+FDEF but standard Java fonts render them as squares. And the mappings here: http://en.wikibooks.org/wiki/Unicode/Character_reference/F000-FFFF just show black squares. Added InterpANSI.Ascii, a convenience "enumeration" containing codes for common ascii characters. 6535452 Dbx console in IDE: still no cursor key support Also forum thread http://forum.java.sun.com/thread.jspa?forumID=852&threadID=5103260 Term will now convert arrow keys per the DtTerm spec: CursorUp ESC [ A CursorDown ESC [ B CursorRight ESC [ C CursorLeft ESC [ D This is done in Term.onCursorkey(). ================================================================================ tag: ivan_22 nobugid remove deprecations - Switch to using setFocusTraversalKeys() as opposed to the deracated isManagingFocus() (in Screen.java). - Use getScreen() as opposed to get getDisplay(). 4921071 printing to the Process Output tab prevents using menus NetBeans has many request processors running at low P1 so a default priority (5?6?) for StreamTerm.OutputMonitor thread will swamp all the RPs if we have a firehose sub-process. Lowering the priority of StreamTerm.OutputMonitor to 1. 4898959 [Debugger Console]: copy/paste via mouse buttons don't work. With 1.4 we now can get a systemSelection in addition to systemClipboard so we can accurately implement X-windows-style selection semantics as follows: SunCopy put Terms selection into systemClipboard (only if non-empty) SunPaste stuff systemClipboard contents into Term buffer selection done put Terms selection into systemSelection clear selection put empty string into systemSelection middle click stuff systemSelection contents into Term buffer The Term API has been extended with pasteFromClipboard() and pasteFromSelection(). The original paste() is now pasteFromClipboard() so NB OutputWindow works like before. Similarly we have copyToClipboard() and copyToSelection() and copy() is copyToClipboard() so NB OutputWindow works like before. Operations with the systemSelection only work if it is available on the host system (For example itis not available Windows). Mouse gestures to stuff the systemSelection _used_ to only work if the autoInsert property is true. That was so that a casual text selection doesn't clobber the clipboard. Now that we have a distinction between the clipboard and selection this property is deprecated and it's setting will be ignored in favor of it always being true. 4898959 [Debugger Console]: copy/paste via mouse buttons don't work. Term used to ignore middle mouse clicks if any mousewheel support was available. Took that test out. See comments in mouseClicked(). 36439 (output window gives ArrayIndexOutOfBounds for some characters) The wcwidth cache is allocated of size Character.MAX_VALUE and indexed by a 'char', so the only way it can get an AOB is Character.MAX_VALUE(\uffff) is passed to it. Fixed by allocating one more cell. Can be easily verified by println'ing a \uffff but it's gotta go through internal execution. 17337 (CTRL-C to copy in Output Window causes it to scroll to bottom) 'keystroke_set' is a collection of KeyStrokes in the form: ks3 = getKeyStroke(VK_C, CTRL_MASK) we use Term.maybeConsume() in keyPressed and keyTyped events. During keyTyped the event->KS mapping gives us ks2 = getKeyStroke((char) ('c'-64), CTRL_MASK) ks2 and ks3 while logically equivalent don't hash to the same so maybeConsume() says yes to ks2 and the Ctrl-C gets passed on. So to detect whether something in 'keystroke_set' needs to be dropped we need to check at keyPress time but take action at keyTyped time. 'passOn' helps us do that. 24824 (Focus problems with splitpane in OW) 4702175 (JScrollBar provide no focus feedback) Issue 24824 pertains mostly to the splitplane confusing things, but the scrollbars getting focus was muddying the waters. Workaround for 4702175 suggets to make the scrollbars not be focusable, so made the horizontal and vertical scrollbars non focusable. The effect of this is Ctrl-Tab will not shift focus to the scrollbars. nobugid (Switched to timed repaint) Per Tims suggestion from issue 28297. I had noticed that pastes (now that I got them working) take an awful long time. A time delay of 20msec does wonders. This should pave the way for simplification of OuputTabTerm as I described in 28297. 36404 (Scrollbars should scroll faster) Until we agree on a common solution changed the rate from 50 to 10 milli-seconds per frame. nobugid (AOOB in Line.insertCharAt()) With InterpANSI run Term under a real pty-based shell and run vi. Go into insert mode. Issue two ^t's and a {. Boom! Line.insertCharAt() could not handle insertions at columns past thelength of the line. Fixed. ================================================================================ tag: ivan_21 issue 24824 Focus problems with splitpane in OW Overrode setEnabled() for Term so it propagates enabledness to sub-components per Aleses request. It's a sensible thing to have in any case. However, I couldn't find the error and setEnabled() code in OW that Ales was talking about. So passing the bug on to Tim who's taken over Ales. regression terminalemulator won't build on JDK < 1.4 I had accidentally left an experimental Clipboard systemSelection = getToolkit().getSystemSelection(); Now it's commented out. ================================================================================ tag: ivan_20 Files: Term.java, Sel.java, Line.java issue 30776 NPE when resizing output window Not enough info, so no action yet. issue 31755 NullPointerException after resizing Output Window The basic problem was that Sel keeps the origin and extent of the selection in unsorted order. Some methods, like paint() and getExtent, setExtent() compensate for this, but adjust() and intersects() didn't. Modified sel.adjust() to take a lastline argument as well. Moved sel.adjust into common area of Term.limit_lines(). nobugid Selection vanishes on resize This used to be done in Term.adjust_lines() to mimic DtTerm, where if you resize so that the current selection ends up going out the window the selection is cancelled. After fixing 31755 it seemed more practical to not nuke the selection (which is how xterm works). issue 31951 Copy to clipboard removing empty lines in output window This was because Line.text would return a "" instead of a "\n" for "empty" lines. This was initially so so that selecting the "empty lines" below the cursor would give "empty" selection strings. But we forego that in order to fix this bug. Turns out xterm also returns newlines for the "empty" lines below the cursor. Issue 21577 addresses the selectability of empty lines below the cursor, but that's orthogonal. Once we can't select these empty lines the fact that they return "" or "\n" per line becomes immaterial. issue 27491 Output window Mouse Pointer Fixed part 2. Mouse pointer is now java.awt.Cursor.TEXT_CURSOR by default. This is consistent with xterm and DtTerm as well. This can always be overriden by using Term.getScreen().setCursor(...); ================================================================================ tag: ivan_19 issue 17644 Executation window cuts off output-window's text java bug 4711314 worked around by adding a repaint to componentResized(). ================================================================================ tag: ivan_18 (OutputWin only) Reversal on invokeAndWait() Issue http://www.netbeans.org/issues/show_bug.cgi?id=25180 Demonstrated several regressions connected with my choice of using invokeAndWait() in OutputTabTerm. David Strupl reveretd by changing invokeNow() to use invokeLater() but neglected to make copies of buffers passed in and forwarded to Term. Also added a quick change flag safe_mode. ================================================================================ tag: ivan_17 Text for bugs: I"m marking this and other NPE related bugs as fixed with my commit tagged ivan_17. For a thorough description read .../terminalemulator/ReleaseNotes.ivan.txt. Since this is a rather radical change I'd rather see new bugs filed as opposed to these being reopened. nobugid slowdown due to accessibility When Assistive Technology latches on to a component various additional property changes get fired. These can be expensive so the usual trick is to only fire them if an AccessibleContext has been requested. However most apps (should) set the accessibleName() and that instantly creates on demand an AccessibleContext. For Term this means that every input character will fire accessible property changes and we don't want thath. We only want to do this if some real AT is latched on to us. So, switched the test to test for an AccessibleText having been doled out. issue 17644 DEFECT P3 PC Ivan@netbeans.org NEW NPE from terminalemulator issue 20412 DEFECT P3 PC Ivan@netbeans.org STAR NPE on org.netbeans.lib. ... issue 24444 DEFECT P3 PC Ivan@netbeans.org STAR NPE changing tab (Editting, ... issue 24728 DEFECT P3 PC Ivan@netbeans.org NEW Random NPE when execute a ... issue 18575 DEFECT P3 PC Ivan@netbeans.org STAR ConcurrentModificationException issue 20430 DEFECT P3 PC Ivan@netbeans.org NEW Deadlock during XTest All of these, I postulate, happen because Term has been used incorrectly. Being that it is a JComponent it's state is only allowed to be modified from the AWT event dispatcher thread. The various NPE and similar problems that arose in the past were unfortunately treated by inserting 'synchronised' all over Term code (mainly because of my incorrect assumption that paints get called on a special repainter thread). There's also one instance of using SwingUtilities.invokeLater() in scrollbar adjustment. With this commit I've reversed the situation. First, all uses of 'synchronized' in Term have been commented out with the following pattern in the comment: "OLD NPE-x". This is to make sure that the _fix_ is fixing the problem and not the leftover synchronized's. In case of disaster the code can be reverted. Next, OutputTabTerm's invocations to Term were routed through SwingUtilities.invokeAndWait() or invokeLater(). They're actually done through little utility functions called invokeNow and invokeLater which do the SwingUtilities.isEventDispatchThread() test. Every stack trace in the above issues has originated from OutputTabTerm so I'm reasonably confident that all the above issues will be addressed by this. invokeNow() is used for InputStream data. invokeLater() is used for actions which come over the dispatcher. Some minor discussion on nbdev raised the issue that invokeAndWait() might induce deadlock. I'd like to argue that this is fine as follows: First, using invokeAndWait() is the rightthing to do for the input. It provides a measure of flow control for the input source and doesn't swamp the AWT event queue. Second, if you use invokeLater(), because you're passing character arrays, these arrays will have to be copied. There's no need for elaborate buffering and queueing since each inner class Runnable which gets created gets it's own copy of the reference to the buffer, but the buffer does need to be copied. Third, as issue 20430 demonstrates, insertion of synchronised is no panacea. We do need queued serialization. Fourth, if we do get deadlocks because of invokeAndWait() we can revisit this question. Regardless, some form of SwingUtilities.invoke has to be used so this fix is in the right direction. Finally, I'm hard-pressed to see how a deadlock can occur. The character input to Term (see below for other "input") comes from an external process or an internal thread. For a deadlock to occur Term code has to vie for a resource that the outputting task is holding. Term is extremely self-contained though. It does not call back into any NB code and therefore should not contend for any resources. (The only exception is the use of debugging println's in internal execution mode which usually cause an infinite recursion). P.S. I actually tried with invokeLater() for a speed comparison. To my surprise I discovered that the text gets all run in and stuff _as if_ ordering gets messed up or the runnables get issued out of order. Instead of pursuing why I took this as further confirmation that invokeAndWait is the right decision. Come to think of it I just used invokeLater and didn't copy my buffers. Analysis of Term state modification In general Term state modifications come from these sources: putChar[s]. This is the main source and it's the responsibility of the caller to call them on the right thread. Various property settings. Happen in constructors or as side-effects of user actions in the gui. So in general they should be safe. All other calls should be carefully scrutinized. Keyboard input Come in on the Event Dispatch thread and usually gets consumed or passed on to a listener. If LineDiscipline() is being used stuff gets echoed but we're still within the Event Dispatch thread. Srollbar notifications, mouse events ... All come in on the Event Dispatch thread. Various mutators in OutputTabTerm are .... Calls from OutputWriter methods of TermOutputWriter. These are the most important source and therefore use invokeAndWait(). setPageMode() historySizeKeeper() Safe. Called from TermOutputWriter Calls from the constructors. Safe. No mutator is going to come in while in a constructor. toString() Unsafe. updatePasteAction <- updateCopyCutAction updateCopyCutAction < activated < TopComponent.componentActivate Safe. activateHyperlink gotoHyperlink invokeJumpListener etc. <- JumpActionPerformer[ActionPerformer].performAction CopyActionPerformer[ActionPerformer].performAction() Unsafe. Called from RequestProcessor. Handled with existing Mutex.EVENT.readAccess. Term.performAction() (only used for PopupAction) boils down to OutputTabTerm.performAction() Unsafe? ... F10 is broken. Already handled with existing Mutex.EVENT.readAccess. selectAll <- actionPerformed setHyperlinkNavigationEnabled <- doClear doClear <- actionPerformed doClear <- topComponentClosed Boils down to OutputTabTerm.actionPerformed() Safe. Called on dispatch thread checkFont <- setSettings setSettings <- propertyChange(PropertyChangeEvent) Safe. Called on dispatch thread Escapes of Term OutputTabTerm provides a getTerm() method, allegedly for testing, but it still may be used so uses need to be scrutinized. ================================================================================ tag: ivan_16 issue 19156 (Not able to navigate to left/right in output window) All code dealing with Ctrl-Tab is gone. If you need to recover it check out code with tag ivan_15. Shortcuts for selection manipulation remain but that is now the subject of http://www.netbeans.org/issues/show_bug.cgi?id=24759 Here is the final result: Action New binding Old Binding Where -------------------------------------------------------------------- Scroll line up Ctrl-UpArrow (1) UpArrow Term Scroll line down Ctrl-DownArrow (1) DownArrow Term Scroll page up PageUp Term Scroll page down PageDown Term Scroll view left Ctrl-PageUp (1) Term Scroll view right Ctrl-PageDown (1) Term Scroll column # No good binding available Next hyperlink Ctrl-T (2) (3) DownArrow OW Prev hyperlink Shift-Ctrl-T (3) UpArrow OW Activate hyperlink Enter|Space Enter|Space OW Activate hyperlink Single-click (4) Single|Double-Click OW+Term Next Error & Activate F12 (3) F12 OW Prev Error & Activate Shift-F12 (3) Shift-F12 OW (1) Conflicts with JLF TabbedPane accelerators. (2) The highlighted errors are best described as hyperlinks hence the generic treatment. (3) If you reach the last (first) link/error next (prev) will not work on the first try and will put out a message in the status bar. One more will cause a wrap then. (4) The original implementation was very confused about single vs double click. So much so that I couldn't characterise it. For example build errors were navigable with a single click, while exception errors had to be double-clicked. ================================================================================ tag: ivan_15 issue 19156 (Not able to navigate to left/right in output window) After discussions with cL on nbui about the merits of Tab vs Ctrl-Tab and a fair amount of work to get Tab to navigate links and Ctrl-Tab to go back to the focus mgr ... turns out the JLF was wrong and that Ctrl-T and Shift-Ctrl-T should be used for link navigation! (this after email exchange with accessibility people at Sun) Since I put a fair amount of work into being able to switch between grabbing Tab, Ctrl-Tab and all I"m commiting this code with all of that code still in AS WELL as Ctrl-T codes so it can be retrieved again and will shortly commit code that elides it. Here are the coupl a notes on that code. CtrlTab. I"m not happy with this decision, so currently you can do it both ways using the property grabFocusKeys() which is set to false by default. The various issues and implementations are discussed in source comments in Term.java and Screen.java both beginning with Dealing with focus traversal ... One of the problematic issues was that Ctrl-Tab (or Ctrl-T) would jump the screen and appear in the term (as boxes) if the OW is not ReadOnly. Solved this problem by introducing setHyperlinkNavigationEnabled() which alters the keysets to consume or ignore Ctrl-T and Ctrl-Shift-T. This stuff kicks in as soon as some exceptions appear so normally these keys go through. Since the keyset is shared had to create two sets. There are still a fair number of overall accelerator issues that still remain but I"ll close this bug and reopen a new one. issue 18733 (Output Window & NotifyException not accessible) One previous "fix" to this which wasn't really was the setting of the name. - Modified Term's accessible context to pass on name setting to the Screen which is the component relevant to accessibility. - USe tab.getName() instead of getName() in OutTermPane.. Otherwise only null names were being passed. The big chunk of work here is the adding of AccessibleText to Screen and all the various support code that's needed for this. All the issues (and there are quite a few) are discussed in the Javadoc comment for Term.getAccessibleContext(). I've unit-tested a fair amount of this functionality, but since I don't have access to actual assistive technology on my Sun box I have no clue if any of this is adequate. I have not yet been able to determine if anything and what needs to be done with the following: - Implement AccessibleState. JTextComponent doesn't do anything special with it, so Term doesn't need to either right? - Implement AccessibleComponent. This one is strange since I can't find any Swing components that implement it! - Term does scrolling, so it would seem like some parts of it need to implement accessible roles of VIEW and SCROLLPANE, but it's not yet clear to me how Assistive Technology would be getting information about this. issue 24460 (Actions "Next/Previous Error" are still enabled after first ... ) I had chopped up checkNextPrevActions() too much. Redid it and it's much simpler now as well. Renamed checkNextPrevActions() to updateNextPrevActions() to bering it in line with other similar functions. ================================================================================ tag: ivan_14 issue 19156 (Not able to navigate to left/right in output window) Not just that but there is precious little keyboard navigation in Term. So this fix attempts to address as much as possible in the whole area of keyboard navigation. There are three rough areas: - Generic keyboard navigation - Error navigation - Selection via the keyboard. All new sequences are documented at the end of this section. OutputTabTerm has a private helper function oldBindings() which checks whether "-Doutput.oldbindings" was set and reverts to older behaviour. Generic keyboard navigation --------------------------- Switched to a set of keyboard bindings that matches the JLF more closely. Moved the few that were implemented in OutputTabTerm to Term itself. Added Term.pageLeft(int n) and Term.pageRight(int n) While testing, ran into and fixed bugs in the horizontal scrolling mechanism whereas the cursor at the end of the line was not scrollable to ... - Term.possiblyHScroll() not uses the cursor position to extend total buffer column size. - ColumnRight()'s limit check was incorrect causing bizarre scrolling behaviour. Implemented the various bindings documented below. Error navigation ---------------- This area itself falls into several sub-parts ... - Treating errors as hyperlinks - "Whole" error hiliting - Hyperlink wrapping - Detecting of errors and utilisation of Term regions - Implementation issues and bwd compatibility. NOTE: A lot of this work really should be done in ActiveTerm but I had to sort of exactly how OW does things before I could attempt to move the functionality down. Treating errors as hyperlinks ............................. OW already tried to treat errors as hyperlinks so all I've done here is (almost) apply JLF rules of hyperlink navigation which are: - Links are traversed using Tab/Shift-Tab. - Links are activated using Space or Enter. - Traversed-to links are denoted using standard swing selection feedback. There are a variety of inherent problems with these as well problems arising in the context of errors. - When showing and navigating java exception dump frames the terminal is not in readonly mode so plain Tab, Enter and Space are out of the question. Ctrl- versions are used instead. - Tab and friends are also focus mgmt keys and it would all work really well if we had not only focus next and prev idioms but also focus up and down idioms. Swing 1.4 is anticipating this but the JLF spec seems to be lagging. (See Component.setFocusTraversalKeys). As it is Ctrl-Tab and Shift-Tab conflict with TabbedPane and OW usurps them by hook and crook. - I don't like selection feedback to mark "current" hyperlinks. If one selects elsewhere the location of the current link gets lost. - There's no specification for denoting the currently activated link. The whole things is also awkward for navigating compiler errors the way we're used to, so F12 and Shift-F12 still work like before. They just imply a next(prev) link combined with activation. "Whole" error hiliting ...................... Previously when navigating from one error you would get: a) A blue/underlined hyperlink like thing on one line. b) A grey character background for the whole error. Since we now use selection hilite to provide navgiation feedback, and since there's no accepted convention for denoting activated links, the grey background is used for denoting the currently activated hyperlink. NOTE: this is a short-term solution. (b) used to be ugly ... it used one per-character backgrounds and as a result had a staircasey look. Changed to full-line background colors. This neccessiated the addition of Term.setRowGlyph() Hyperlink wrapping .................. At the last (first) error a next (prev) action will have no effect but putting a message into the status bar, but one more action will cause the navigation wrap around to the first (last) error. This neccessiated the addition of ActiveRegion.lastChild() to complement ActiveRegion.firstChild() Detecting of errors and utilisation of Term regions ................................................... It also turned out that the way regions were created for compiler errors was very different from java exceptions. In the compilers case the boundries are know ahead of time and regions can be created. In the case of stack errors no regions are really created, the line is pattern-matched _after_ it's been sent to Term and an independent hit targeting scheme, based on the variable 'links', is used. So I switched this to on-the-fly pattern recognition with a small state machine. This way java exception hyperlinks are now treated more like regular errors and the whole 'links' mechanism can be eliminated in the future. As a result exception error navigation is a bit more predictable. This whole area is still waay too ad-hoc and brittle. Implementation issues and bwd compatibility ........................................... Created a parallel set of variables and routines in OutputTabTerm to satisfy the new key bindings and semantics. old new ..................................................... nextPrevJump() nextHyperlink | prevHyperlink changeCurrentregion() gotoHyperlink() activateHyperlink() currentRegion currentHyperlink oldBIndings() (mentioned above) plays it's biggest role here. The hardest part here was battling the focus manager for control over Tab, Ctrl-Tab, Shift-Tab. See the comments with the heading "Dealing with focus traversal". Selection via the keyboard -------------------------- This will be done later. Action New binding Old Binding Where -------------------------------------------------------------------- Scroll line up Ctrl-UpArrow (1) UpArrow Term Scroll line down Ctrl-DownArrow (1) DownArrow Term Scroll page up PageUp Term Scroll page down PageDown Term Scroll view left Ctrl-PageUp (1) Term Scroll view right Ctrl-PageDown (1) Term Scroll column # No good binding available Next hyperlink Ctrl-Tab (2) (3) (5) DownArrow OW Prev hyperlink Shift-Tab (4) (5) UpArrow OW Activate hyperlink Ctrl-Enter|Ctrl-Space (3) Enter|Space OW Activate hyperlink Single-click (6) Single|Double-Click OW+Term Next Error & Activate F12 (5) F12 OW Prev Error & Activate Shift-F12 (5) Shift-F12 OW (1) Conflicts with JLF TabbedPane accelerators. (2) The highlighted errors are best described as hyperlinks hence the generic treatment. (3) The general rule is that if there's no textual input that Tab should be used, a Ctrl-Tab is for cases where we have textual input because there Tab is meaningful. However it's very hard for users of Netbeans to know when an output window pane is in "readonly" mode, so we basically say always use Ctrl-ed variations so you don't have to think about it. (4) Technically Shift- is a direction reverser, so if the forward direction is Ctrl-Tab then the bwd direction should be Ctrl-Shift-Tab ... except that youhave to have fingers of a Martian to be able to do that, so I've stuck to Shift-Tab. (5) If you reach the last (first) link/error next (prev) will not work on the first try and will put out a message in the status bar. One more will cause a wrap then. (6) The original implementation was very confused about single vs double click. So much so that I couldn't characterise it. for example buld errors were navigable with a single click, while exception errors had to be double-clicked. issue 18733 (Output Window & NotifyException not accessible) Phase-I: Term is accessible with role PANEL. It's just a generic container. ScrollWrapper is accessible with role PANEL. It is just a wrapper around the horizontal scrollbar to help in it's placement. Screen is accessible with role SWING_COMPONENT. It should by rights be TEXT but it's very tricky to implement one dimensional caret coordinates in a 2D text widget. So, that's for later. These "fixes" deal with the accessibility issue only very superficially. That is, all JComponents that go into Term implement Accessible and return a description and a reasonable, but not neccessarily useful, role. More needs to be done if these components are to be actually usable. - Screen needs to implement AccessibleText. - Term combines a scrollview and a view into one widget while Swing accessibility expects these to be independent. I"m not sure how to address this. One way would be to create dummy widgets that declare the roles SCROLL_PANE and VIEW just so they can work with accessibility. nobugid (OutputTabTerm timer was never really effective) While working on error/hyperlink navigation I noticed that Term's refreshEnabled property was always on. This in effect makes the whole timer approach introduced by rmatous not work. The way that was supposed to work is: 1) issue Term.setRefreshEnabled(false) 2) process and send chars to term and trigger timer 3) when timer fires cause a repaint. It looks like though that step (1) was never done. This is extremely puzzling since at the time (cvs log) rmatous turned on timer on 2001/09/03 there was ample proof that the timer helped. My suspicion is that at around the same time, this happenned: rmatous fixed CopyMaker on 2001/08/28 13:09:12 so Term gets buffered input and the improvement was attributed to the wrong fix. In any case I've introduced step (1) and measured an increase in speed. Also, the code in repaintTimer() where the refreshEnabled property gets saved, set and restored is all redundant, since the earlier call to flush() does all this anyway. ================================================================================ tag: ivan_13 - performance - reduce Interp footprint This was based on an observation made by Tor that each Interp ends up creating redundant copies of it's tste tables. All Interps now have a static inner class InterpType which owns the state transition tables and defines the actions.. Multiple instances of Interps of the same type share InterpTypes. Since the state transition actions are now implemented in the InterpType, they need to receive an instance of an Interp whose state they will be modifying. This is passed as an AbstractInterp. Occasionally the passed-in interp has to be cast to the appropriate subclass of AbstractInterp. In order to reduce the number of these casts moved number parsing mgmt from InterpANSI to AbstractInterp. Some Interp subclasses achieved their means by modifying their state vectors! Since the vectors are now shared that won't do, so a more appropriate state stack was introduced into InterpDumb. The stack is cleared in reset(). Files: AbstractInterp.java InterpANSI.java InterpDtTerm.java InterpDumb.java - performance - user cheaper Interps by default. Jesse (I think) pointed out that NB in general has no use for ANSI emulation, and that Term should by default use a "dumb" terminal. This should reduce the # of classes that get loaded in. This happens in the initialization of 'private Term.interp'. - I18N This addresses the following issues: 15333 Cursor isn't on end of text after using CTRL+C [V, ...] 19570 I18N - The characters of the error message are overlaped. Basically Term can now properly handle non-latin characters, like Japanese. These issues were realy only the tip of the iceberg. Term did not really work with japanese until now. The following work: - Proper cursor position. - Proper handling of line wrapping and backspacing over it (for when horizontallyScrollable is false). This is important for the proper working of Solaris 'vi' in the ja locale. - Sane reaction to ANSI terminal control escapes. - Selection works. - Active regions work. - Backspace, TAB etc. work. There are two big parts to this - Rendering characters in grid/cellular fashion. The book Creating Worldwide software (second edition) (Prentice Hall) Tuthill & Smallberg discusses (on p98) how some characters might be double width and presents 'wcswidth(3)' and 'wcwidth(3)' to return the _display_ width of a given character. The underlying assumption here is that fixed width fonts are actually quantized width fonts. This doesn't seem to be the case for Java fonts. For example the default ja font I get has 7pixel wide latin characters and 12 pixel wide japanese characters. What to do? Write our own 'wcwidth' that uses Font.charWidth() and rounds it up to a multiple of the width of the latin char sub-set. But that's not enough. Graphics.drawString() will still advance each glyph by the original width of the font, not our rounded-up value. One solution is then to use a drawString() per character. The adopted solution is instead to use GlyphVector's and Graphics2d.drawGlyphVector() as used in Term.myDrawChars() and Term.massage_glyphs(). Despite the hairiness of the code there it turns out to be faster than a drawString() per char by a good margin. - Accounting for the difference in Buffer vs Screen coordinates. (A reading of the main javadoc comment for class Term would help understand the rest of this). So now we have a situation where a Line holds characters whose positions are not neccessarily in a 1-1 correspondence with their cell positions. Mappings are provided in both directions via Line.bufToCell() and Line.cellToBuf(). They are used in the existing buffer to view coordinate xform functions (which for example map a screen position to a character for the purpose of selection). A variety of other locations had to be adjusted to use these for proper operation. The driving algorithm for choosing what needs attention was occurances of st.cursor.col since the cursor is in cell coordinates. These function aren't "cheap" because they count from the beginning of the line. Cacheing the values is impractical for the following reasons: - You need to cache each mapping since they are used with equal frequency. - Because Term allows horizontal scrolling a line can potentially be very long. The index into a line therefore will range from 0 to Integer.MAX_VALUE. This means a cache of short's won't do. - So now we're talking a fair amount of memory that is probably not justifiable unless we come up with a way to quickly dispose of the caches. Cache invalidation is always a tricky problem, but I found out something else. For a while I installed a wcwidth() cache per line (In retrospect having wcwidth() manage the cache of course made a lot more sense) but along the way I discovered that the per-line cache gets invalidated quite often. In effect the cache wouldn't have been helpful. There are some pattern that could use improvement, a bufToCell immediately followed by a cellToBuf, or a bufToCell(x) followed by bufToCell(x+n). These could be collapsed into specialized functions. Some neccesssary fallout from all of this ... - MyFontMetrics.wcwidth() is expensive given the number of times it gets called so the resultant width is cached. The cache is indexed by the char so is naturally Character.MAX_VALUE big. Not a big chunk of memory in the big scheme of things but it can add up if you have many Term instances. So there's a pool of them indexed by FontMetrics. It's unfortunately trickier than you'd think. See the opening comment in class MyFontMetrics for more info. - LineDiscipline had to be adjusted to do something reasonable with backspaces in line buffered mode. There's a big comment in LineDiscipline.sendChar(). To do this it reuires to have a back-pointer to the Term so we now have StreamTerm.setTerm() which is used in Term.pushStream(). - Dealing with double-width is expensive so we don't want to compromise speed in 8-bit locales. One way to deal with this is to query the "file.encoding" property but I found that it's value is very variable from Java release to Java release and probably from platform to platform. In 1.4 class Charset is supposed to deal with this but we're not ready for that switch yet. What I opted for is having MyFontMetrics.wcwidth check for variation from an initial width and set a flag (multiCell) on the first deviation. Various parts of the code then check MyFontMetrics.isMultiCell(). So, for example, the painting code now instead of calling Graphics.drawChars() will call Term.myDrawChars() which will based on this flag do the expensive or the cheap thing. A note on ANSI emulation vs double-width characters. The ANSI standard doesn't talk abut double-width characters! So dealing with them is AFAIK up to individual vendors. I've followed Solaris'es DtTerm behaviour and spent a fair amount of time making sure that Solaris vi (which can excellently edit wide-char files under DtTerm) works under Term. - bug: Junk characters inserted on character insert. Occasionally when charactes are shuffled in the Buffer, usually under vi, junk (usually 0) characters gets inserted into the Buffer instead of ' ' (ASCII SP)'s. These show up as "squares". Fixed in Line.insertCharAt(), Line.setCharAt() and Line.clearToEndFrom() Prior to JDK 1.4 ascii 0 was rendered by Swing as a blank. But under 1.4 this problem is a whole lot more visible. - bug: Pathologically slow horizontal scrolling on long lines Was attempting to render all characters even those that are not in view. Fixed in Term.paint_line_new(). - bug: Cursor gets drawn on the glyph gutter on horizontal scroll. Fixed by adding an additional check in paint_cursor(); - deprecated: Term.goTo(Coord) Use Term.setCursorCoord(Coord) which matches getCursorCoord(). - Misc: public->private Buffer.visible_cols + Buffer.visibleCols() + StreamTerm.BUFSZ (instead of hard-coded 1024) + collection of statistics on linefeeds - Term.paint_line_old() // dead code + Term.charWidth(char) // See section above on I18N. ================================================================================ tag: release33 - accessibility Term now implements Accessible and returns an accessible context. The "accessible description" is set. The "accessible name" is set from OW to be the same as the tab name. - performance - Added Term.setKeyStrokeSet() in order to allow sharing of, sometimes large, sets. Added code to OW to take advantage of this. See 'updateKeyStrokeSet()' and 'getCommonKeyStrokeSet()'. This code also tracks changes to the global keymap so that Term will now pass through newly added keyboard accelerators. ================================================================================ back to main 3.3. trunk - bug: Missing ANSI escape sequence handling: - ESC [ 4 h set insert/overstrike mode - ESC [ 4 l reset insert/overstrike mode - ESC [ 4; 17 r margin control (used by vim for sub-windows) - ESC c full reset - ESC [ ! p soft reset - ESC [ n status report. (used by unix command 'resize') This also required cursor positioning ('H') to clip as opposed to ignore out of bounds row and column settings. - bug: Exception when running "xemacs -nw" under pty-based Term xemacs has a propensity to send cursor motion directives with rows that exceed the boundaries and the checks in Ops.op_cm() were inadequate. - issue 16010 (Autoscrolling behavior of terminal not ideal) Added a new properties boolean scrollOnOutput boolean trackCursor to complement the existing scrollOnInput. When 'scrollOnOutput' is set to 'false', you can use the scrollbar to look at some text higher up w/o it moving from underneath you as more output is produced. This feasture is "smart" in the sense that if the cursor is visible (i.e. the user hasn't scrolled away from the cursor) Term will scroll to track the cursor even if scrollOnOutput is set to false. However the smarts only kick in if 'trackCursor' is set to true. Adjsuted in netbeans/core/output/OutputTabTerm.java as well by adding term.setScrollOnOutput( false ) etc. - new: added the following under CVS control: ReleaseNotes.ivan.txt This file. build.xml For localized builds properties.html Used by func_spec.html interpreter.html Used by func_spec.html func_spec.html Evolution from proposal.3.html proposal.3.html Original proposal as it appeared on the NB site. ================================================================================ tag: term_aug2001_ivan_12 - optimization Text with attributes (fg color, bg color, underline, active, etc) used to be rendered one character at a time and rather slowly. Switched to run-length based rendering where runs of characters with identical attributes are rendered together. Depending on the density of attributed text this has produced a *2 to *10 speedup in view painting. The function implementing this is Term.paint_line_new(). paint_line_old() has been kept around just in case. ================================================================================ tag: term_aug2001_ivan_11 - bug http://openide.netbeans.org/issues/show_bug.cgi?id=16027: Missing/awkward selection access functionality + public void paste() + public void copy() + property selectionExtent is now bound and fireProprtyChanged() is issued when selection changes. - bug http://openide.netbeans.org/issues/show_bug.cgi?id=15953 Exceptions when resized on selection. Fixed by adjusting selection in Term.adjust_lines. This is a stop-gap fix. All the selection adjustment code that uses Sel.intersection should be moved into Buffer methods. - BG line stripe support (Term.setGlyph()) - Eliminated interference between BG line strips and selection. Did this by drawing the BG stripes before the selection in Term.do_paint(). - The stripes were being drawn in the gutters which made it look funny. Now they're being draw only in the text area. - bug BG color of characters would override Swing style selection. Fixed by having paint_line() not bother with BG rectangles if the character falls into a selection and always set the character FG color to the default fg. Added Extent.intersects(int, int) to help with this. - bug From Ales and NB testers: > 1. When you run following simple program: > public class Out { > public static void main (String args[]) { > System.out.println("123"); > } > } > > and then you try to select more lines in OW, it looks strange. > - selection area ended somewhere in the middle of OW width > - when you go (by selecting) behind OW right bound, squares will > displayed in start of line (see squares.gif) > - when you run this program second time, everything looks fine > Reproduced and fixed. There are two parts to the problem: 1) The selection is halfway through. 2) When you select and drag _out_ of the window you get the funny view. In both cases it had to do with (different) variables going below 0. Now, after a resize, line selection should extend all the way and the drag right should work right. - bug From Ales and NB testers: > 2. Sometimes (when playing with selection) we got NPE in > Term$Scroller.extend method, on 1125 line > I found out, that sel.sel_extent is null from some reason > This will happen if the selection "vanishes" while auto-scrolling is on. The way I reproduced it was by having a program trickle out a line of junk , once per second, then while it's running and scrolling select some test and drag up and out of the OW thereby starting the auto-scroller thread. Eventually the selection will go out of buffer history (If you have a short history or no anchors) and I got the same exception. I fixed it in Term.Scroller.extend() by checking for a null selection extent. That whole block of code is synchronized with Term.this which is the main lock for everything. - Eliminated lots of OLD code. ================================================================================ tag: term_aug2001_ivan_10 + Term.columnLeft(int n) + Term.columnRight(int n) These are analogs of lineUp/lineDown() for horizontal scrolling. - auto-scrolling of dragged selections now works for both directions. - auto-horizontal-scrolling to keep cursor in view. Only on input. This is tricky since because of non-local echoing the cursor position is not known on a keystroke. See comment above hscrollReset() for a detailed explanation of the design. - bug Attributed text would not get rendered correctly on horizontal scroll. Fixed. ================================================================================ tag: term_aug2001_ivan_9 + Horizontal scrollbars. Use Term.setHorizontallyScrollable(boolean) to enable/disable it. Still needs auto scrolling and probably some other smarts. ================================================================================ tag: term_aug2001_ivan_8 - Sprinkled NOI18N's all over. - Commented out debugging println's ================================================================================ tag: term_aug2001_ivan_7 - bug 15365 (View does't folow cursor - when writing) Fixed by adding a boolean property scrollOnInput() which by default is set to true. - Added boolean property readOnly. When true, keystroke events and paste's are ignored. ================================================================================ tag: term_aug2001_ivan_6 + Term.pageUp(int n) + Term.pageDown(int n) + Term.lineUp(int n) + Term.lineDown(int n) Scroll the view 'n' pages/lines up/down. Doing key bindings for these is up to the Term client. - Implemented selection auto-scrolling for when mouse pointer moves out of the view. This is based on feedback from Chris Ledantec on nbui. - Implemented SHIFT-left-click as a selection extension mechanism, per JLF. Actually it is slightly off in that it implements extension (what you get when you drag) not addition semantics. - bug: Term.possiblyNormalize() would occasionally cause a null Line exception. Fixed. - bug: Tab expansion add's a '\0' into the buffer. On Windows this shows up as as the Microsoft "unprintable character" square. Easily fixed in Term.OpsImpl.op_tab() by switching the ' ' character addition and the st.cursor.col++; ================================================================================ tag: term_aug2001_ivan_5 - bug: The common "null Line" bug fixed. Turns out I was using "synchronized(this)" in an inner class of Term. - bug: minor bugs having to do with null'ed selections fixed. I think these only surfaced when I implemented absolute coordinates. - Absolute Coordinates are in. What this means is that you no longer need to anchor text in order for Regions to work. The way this is accomplished is by having each line get an ever growing number and have Coord and ActiveRegions use these numbers for rows. implementation details: ---------------------- To separate such "absolute" coordinates from the original buffer coordinates, instroduced package private class BCoord. Coord and BCoord can can be converted to each other. The conversion is based on 'Term.firsta' which is the absolute line number of the first line in thebufer. Just like Term.st.firstx is the buffer coordinate of the first line visible in the view. Similarly a BExtent was provided in parallel with Extent. Serious attention was payed to the possibility of 'firsta' wrapping around it's 32bit range. Ideally this could be dealt with by using unsigned numbers and modulo arithmetic, but Java doesn't have unsigned. Instead the variable is explicitly checked against Term.modulo = Integer.MAX_VALUE/2. The 2 is for good measure. Term.modulo can be artificially reduced to a small number in order to verify that selection and ActiveRegion relocation works correctly. ActiveRegions that go out of history periodically get culled. This is controlled using Term.cull_frequency. - debugging aids: - Added a "debug gutter" Which can be used to print out view, buffer or absolute row coordinates. - Added Buffer.lock and Buffer.ck_lock() which helped me find race conditions. The code is commented out. ================================================================================ tag: term_aug2001_ivan_4 - preparatory work for processing of ANSI escape sequence ESC [ ; r ================================================================================ tag: term_aug2001_ivan_3 - bug: ActiveRegion.setSelectable(boolean) always set property to true. Fixed. - bug: After the flush cleanup in term_aug2001_ivan_2 LineDispcipline's eches of characters wouldn't show up until after a newline if the refreshEnabled property was off. Fixed by adding additional flushes. + Term.requestFocus() ... was missing. Added. + ActiveTerm.cancelRegion(). This is supposed to help with hiliting regions that are detected only after the fact .. that is you aggressively call beginRegion(), like at the beginning of a line, and if at the end of the line it turns out not to be useful you can cancel it. ================================================================================ tag: term_aug2001_ivan_2 - Fixed a bug where the Sun Copy key wouldn't copy things. + Term.flush(), TermStream.flush() See JavaDoc for Term.flush() for more info. - Internal change to repaint(). Term.repaint() used to always check the refreshEnabled attribute, which would cause things like scrolling or attribute changes not take immediate effect. So now we have possibly_repaint() which checks the attribute and repaint() which doesn't. Only putChar() and putChars() use possibly_repaint(). Made repaint() and possibly_repaint() be protected. ================================================================================ tag: term_aug2001_ivan_1 - AbstractInterp, InterpDumb, InterpANSI made public - class ActiveTerm now inherits from class StreamTerm instead of class Term + ActiveRegion ActiveRegion.firstChild() + ActiveRegion ActiveRegion.getNextSibling() + ActiveRegion ActiveRegion.getPreviousSibling() To be used as follows: ActiveTerm at; r = at.regionManager().root().firstChild(); r = r.getNextSibling() ... You should be able to use Regions instead of Coords to index actions (see OutputTabTerm., new ActiveTermListener. + int Ops.op_get_width() + int Ops.op_get_column() To help with formatting and wrapping text in a Term - The OutputStream 'pin' parameter to StreamTerm.connect() is now optional. This makes sense for output only situations. + void Term.setTabSize(int tab_size) + int Term.getTabSize() Per NB folks' request. You should be able to get rid of expandTabs() in org/netbeans/core/output/OutputTabTerm.java.