CHANGELOG:
v1.20.0
  - Changed an internal test to make it comply with an upcoming R change.
  - Updated various URLs in the vignette to comply with ever-evolving CRAN
    rules.
  - Converted various network related functions to generics, making it easier
    to support e.g. networkLite objects.
v1.19.0
  - Non-visible and documentation changes for CRAN compliance.
  - Internal argument handling changes; should generally be invisible, but
    may in some complex cases avoid problems when calling network().
  - Added more internal checks to edge allocation and handling.  These should
    not be visible for normal use, but should catch/avoid errors due to 
    improper use (e.g., trying to make/use network objects with malformed
    edges).  This is not expected to break any working code, but if you
    have been doing things you shouldn't have been doing (e.g., edges
    without both inl and outl elements), then you might find that you need
    to fix things.  (This usually takes effort, so you are not likely to
    have done it by accident.)
  - Bolstered add.edge and add.edges against improper edge.check arguments.
  - Made get.inducedSubgraph() and permute.vertexIDs() generic.
  - Replaced now non-functional Pajek URLs with Internet Archive versions.
v1.18.2
  - Non-visible changes for CRAN compliance.
v1.18.1
  - Non-visible changes for CRAN compliance.
  - as.network.data.frame() now handles deleted edges correctly.
  - as.data.frame.network() can now mimic as_tibble_network() behavior.
v1.18.0
  - Many network methods are now S3 generics (to facilitate objects like networkLite).
  - get.dyads.eids now has an na.omit argument.
  - Minor bug fixes.
v1.17.2
  - Deactivated tests to appease CRAN.
  - Modified various protection calls in a hopefully more conservative manner.  This should not produce user-visible changes, but will be cleaner on rchk.
v1.17.1
  - Coercion from an edgelist could fail with a single edge when redundant edges were being removed.
  - Zero-row matrices are now assumed to be empty edgelists when not explicitly specified.  (Before, they just failed.)
v1.17
  - Michal Bojanowski, Pavel N. Krivitsky, Chad Klumb, and Brendan Knapp have been added as contributors.
  - summary.character() and print.summary.character() are no longer exported as it interfered with base::summary.data.frame().
  - Added functions as.network.data.frame and as.data.frame.network for converting between data.frames and networks.
  - as.tibble.network and as_tibble.network now optionally return edge IDs, controlled with the store.eid argument
  - Fixes to read.paj().
  - Various testthat improvements.
  - Bugs fixed in mixingmatrix.network().
  - Updated tests for compatibility with R 4.0
  - Converted is.bipartite, is.directed, get.edge.attribute, get.edge.value, get.network.attribute, list.edge.attributes, list.network.attributes, set.vertex.attribute, set.edge.attribute, set.edge.value, set.network.attribute, delete.vertex.attribute, delete.edge.attribute, delete.network.attribute, has.vertex.attribute, has.edge.attribute, and has.network.attribute to generics for greater extensibility.
  - Using an out-of-bounds index in nw[i,j] and nw[i,j]<- now produces a sensible error message (i.e., same one as for a matrix).
  - mixingmatrix() now produces output inheriting from "table" rather than a list; its printing, handling of missing attributes, bipartite networks, and other scenarios has been improved.
  - as.mixingmatrix() has been added.
  - network now uses statnet::statnetStartupMessage().
  - Updated constructor code per Kurt Hornik's suggestion to future proof versus upcoming sapply() changes.
  - network.edgelist() now automagically prunes loops and redundant edges for networks that should not have them; previously, the function trusted the user not to supply such things (and would go ahead and make edges if asked to do so).
v1.16
  - Converted get.vertex.attribute and list.vertex.attributes to generics for greater extensibility.
  - Updated tests to fix a color-related issue.
  - Switched explicit class matching checks to inherits().  This avoids a compatibility problem with future R releases (and facilitates further class extensions).
v1.15
  - made na.omit a non-required arguments S3 method generics for network.edgecount and network.dyadcount for exensibility by downstream packages
v1.14 Fixes & Features
   - revamped the network C API, which had been broken by earlier regrestration changes.  Users making use of the C API will need to update their networkapi.c and networkapi.h files, but no other changes need to be made to existing code.
   - Fixed an un-handled case for plot edge.lwd expansion when called from other contexts. It can now replicate a single numeric value even if scaling is not otherwise included.
   - network now has an as_tibble() method to convert it to a tibble edgelist with specified edge attributes as columns or to a tibble vertex list with vertex attributes as columns.
   - as.edgelist() can return a tibble as well.
   - network print() methods now handle formula network attributes intelligently.
   - mixingmatrix() is now a generic, with a mixingmatrix.network() method.
   - Improved handling for NAs in network subsetting assignment operation.
   - A number of functions (network.edge.count(), network.dyadcount(), network.naedgecount(), and network.size()) are now generics.
   - Assigning a vector (as opposed to a matrix) to %e% (edge attribute) now sets the edge attribute of corresponding edges (in the same order as the internal representation).
   - Various changes made to satisfy CRAN's ever-shifting code requirements.
v1.13 Fixes & Features
   - added some CRAN requested imports to NAMESPACE 
   - added get.dyads.eids function to return sets of edge ids for vectors of tails and heads vertex ids for non-multiplex networks
   - copied as.edgelist functions from ergm, they serve as wrapper for as.matrix.edgelist.network that enforces sorting and adds an 'edgelist' class. 
   - added a has.edges() function to determine if a vertex has any edges (is not an isolate)
   - updated network.dyadcount to account for loops and directed bipartite networks
   - added optional opacity paramter to as.color, and new docs for as.color.
   - rewrote read.paj() parser for Pajek formatted files, now supports timing and edge attributes, fixes error for empty network
   - added Suggests dependency for testthat for testing
v1.12 Fixes & Features
   - fix error with as.matrix.network.edgelist when as.sna.edgelist=TRUE and network has 0 edges
   - updated citation generation code
   - NAMESPACE file updated to explicitly register S3 methods (and some pseduo S3 methods as requested by CRAN check)
   - changing the vertex.sides argument to plot.network now produces polygons of equal area instead of equal radius.
   - plot.network can now plot labels on curved edges
   - refactored plot.network so that much of the argument expansion is handled by a public method plotArgs.network so that it can be accessed by depending packages
v1.11 Fixes & Features
   - minor change to layout.c to fix UBSAN warning flagged by CRAN
   - minor changes to print.network to give more informative info about some network attributes
   - get.edge.attribute now has C-level backend
   - get.edge.attribute and get.edge.value now have arguments to avoid returning attribute values from 'missing' edges (na.omit) and unset/deleted edges (deleted.edges.omit) and edges with no value assigned for the attribute (null.na). 
   - network now includes an explicit list of exported functions in NAMESPACE
   - the following 'internal' functions have been removed from NAMESPACE: networkOperatorSetup, switchArcDirection, readAndVectorizeLine, read.paj.simplify
   - network now follows R conventions for exporting some of its C-level API functions for use by other packages
   - get.inducedSubgraph() now has an optional 'eid' argument for inducing a network including specified edges and their incident vertices
   - fixed bug in as.network.matrix that caused it to ignore the na.rm argument.
   - added patch to which.matrix. type to handle NAs submitted by Philip Leifeld
   - fixed na.rm argument to as.network.matrix
   - changed plot function to force lazy evaluation of displaylabels=!missing(label) argument before value of label is changed
v1.10 Fixes
   - fixed vertex label plotting bug when vertices have the same positions
   - added dimension checks and more informative error for as.network.matrix for case when it assumes a square edgelist is an adjacency matrix
   - added dependency on statnet.common library to facilitate some development testing
   - The 'modify in place' syntax for network functions that modify their arguments now works correctly when the network to be modified is an element of a list
   - Modified plot.network.default so that if a single value is passed in that matches with a vertex attribute name, the value of the attribute will be used (to make it consistent with the rest of the plot arguments)
   - Fix to C-level stack protection bug in delete.vertices that could cause crashes under some conditions
   - removed $<-.network assignment operator so that dispatch will occur via R internals. 
   - Fix stack protection overflow in set.edge.value. Also check for correct dimension on value argument for set.edge.value.
   - Minor documentation fixes
v1.9 Features & Fixes
  New Features
    - minor tweak to print.network to not show summary information for net.obs.period and .active attributes when called from network dynamic
    - add.vertices, add.edge, and add.edges have been converted to S3 methods
    - set.vertex.attributes and set.edge.attributes can now modify a list of on a network attributes with a single call
    - modified all of the set and add methods to remove the x$RisTooLazy?<-NULL copy-inducing kludge in favor of x<-duplicate(x) in the C version of the function. Seems to give moderate speed improvement for large networks
  Fixes
    - fixed a c-level bug in get.edge.ids that was giving bad values on undirected networks with loops.
    - fixed a bug where [<-.network assignment operator fails to add edge in non-directed loops case

v1.8 Fixes & Features
  New Features
    - now supports bipartite networks with zero-sized first mode. (bipartite=0 no longer means the same thing as bipartite=FALSE, and is.bipartite() reflects this)
    - added vertex.lwd attribute to plot.network for controlling line widths around vertices
    - added basic plotting for edge labels (new plot.network arguments: edge.label, edge.label.cex, edge.label.col)
    - added valid.eids function to return the ids of non-NULL edges in a network
    - added a loading.attributes man page with more examples of attaching vertex and edge attributes
    - print.network now lists edge attributes if there are less than 1k edges, otherwise a notice that attribute names are not shown (listing edge attributes is expensive)
  Bug Fixes
    - permute.vertexIDs was triggering a protection stack overflow on large graphs.
    - fixes for the undirected network subtraction operator and ! operator bug. resolution involved forcing a i<j ordering for comparisons of undirected networks.
    - fixed network.edgelist and as.network.matrix so that it is possible to construct networks using a matrix or dataframe in which the first 2 cols give tail and head ids, and remaining cols define attributes as docs state.
    - fixed network.bipartite constructor so that it can correctly handle edge weights when building network from bipartite matrix. Also added automatic cast to matrix so it will work with data.frame argument.
    - fixed  bug in how is.color() processed NA values that mangled colors in plots in some situations.

v1.7.2 New Features/Changes/Bug Fixes
  New Features:
    - networks with zero vertices (zero-order networks) are now supported. Some functions now give explicit errors for networks with size zero because behavior is not well defined.
  Changes
    - package is now hosted as part of the statnet project
    - get.edge.attribute will now silently call get.edge.value if accidentally called with a network instead of net$mel.
    - some minor additions to documentation
    - original JStatSoftware article included as package vignette
    - package now requires the sna package, to provide some functions needed for network layouts
  Bug Fixes
    - fixed bug in add vertices when adding to second mode of bipartite network
    - fixed deprecation warning for R v2.15.3
    - Fixed bug where replacement operator can add multiple edges to pair of vertices

v1.7-1 Bug Fixes
  Bug Fixes:
    - .onLoad's startup message code caused excitingly bad behavior with recent
      versions of R. [CTB]
    - network.adjacency, network.edgelist, etc. were no longer modifying their
      arguments in place, and have had their correct behavior restored.  [CTB]
    - delete.vertices was building up too many protection stack calls, which
      could cause issues in some cases.  [CTB, submitted by Adrian Albert]

v1.7 New Features/Changes/Bug Fixes
  New Features:
    - as.matrix.network.edgelist now supports an argument, as.sna.edgelist,
      that will return a fully compliant sna edgelist representation of the
      network.  For sna versions 2.3 and later, this will among other things
      allow missing data information to be automagically retained when passing
      network objects to sna functions.  (Previous to this, missing edges were
      stripped on conversion.)  [CTB]
  Changes:
    - print.summary.network no longer dumps excessive information when printing
      edge lists.  [CTB]
    - Built-in data objects are now compressed (not that it made much
      difference, but CRAN was complaining).  This also requires R version
      2.10 or later (previously required R>=2.0).  [CTB]
  Bug Fixes:
    - The network summary and print methods could in some cases fail if called
      on a network with non-trivial graph-level attributes.  [CTB, submitted by 
      Zack Almquist; closely related bug fixed by MSH]
    - If called with ignore.eval=T but no specified names.eval, as.network
      and friends generated an object with mislabeled edge attributes; note
      that a warning is still given with such a call, since it is unwise not
      to specify attribute names.  [CTB, submitted by DH]

v1.6 Changes/Bug Fixes
  Changes:
    - In-place modification methods now use draconian measures to force 
      evaluation of their arguments prior to modification, and overwrite
      non-anonymous arguments in their original frame on exit.  This is a kludge
      to cope with R's lazy evaluation strategy, which does not play well
      with in-place modification of arguments.
  Bug Fixes:
    - Per the above, semantics for in-place modification of objects should now
      operate correctly (was conflicting with R's lazy evaluation scheme,
      sometimes resulting in bizarre behavior).  [CTB, submitted by Ronggui
      Huang]
    - plot.network was failing when edge.col was called with single color terms.
      [CTB, submitted by Philip Leifeld]
    - network.layout.fruchtermanreingold failed when called with a network
      having exactly one edge.  [DRH]

v1.5-1 Minor Changes/Bug Fixes
  Changes:
    - The mixingmatrix function has been moved from ergm to network (currently
      as an undocumented internal function).  [CTB, but original function is
      due to MSH]
  Bug fixes:
    - plot.network generated an error in some cases when edge.curv was set.
      [CTB, submitted by Brian Ripley]
    - summary.network generated an error if mixingmatrices=TRUE was set and
      ergm was not loaded.  (The mixingmatrix function has now been moved into
      the network package.)  [CTB, submitted by Ronggui Huang]

v1.5 New Functions, New Features, Changes, and Bug Fixes
  New Functions:
    - get.inducedSubgraph: return induced subgraphs and edge cuts from a 
      network object.  A new operator, %s%, has been introduced to simplify
      this process.  [CTB]
  New Features:
    - add.vertices now supports adding vertices to the first mode of a 
      bipartite network (default behavior is unchanged).  [CTB]
    - as.matrix.network.adjacency and as.sociomatrix now support an optional
      argument to force bipartite graphs to be returned in full one-mode 
      adjacency form (rather than the current, two-mode default).  [CTB]
    - print.network and print.summary.network now support an argument to allow
      suppression of matrix output (helpful for very large graphs).  [CTB]
    - network.layout.fruchtermanreingold now uses a cell-based acceleration
      scheme for large graphs, which can be adjusted using layout.par; it also
      operates entirely on edgelists, and is no longer O(N^2) in typical
      applications.
    - Network coercion methods now recognize/use sna edgelist attributes
      (even if the matrix is not actually an sna edgelist!) automagically.
      Coercion to edgelist form also sets the sna edgelist attributes.  In
      addition to being useful for interoperability, this now makes it easier
      to import network data in edgelist form (previously, one had to be 
      careful about setting vertex sizes, which could only be crudely inferred
      from the edgelist matrix itself).  [CTB]
  Changes: 
    - By very popular demand, network now uses R name spaces.  [Credit to
      Michal Bojanowski for convincing us to make the leap]
    - Also by popular demand, print.network no longer displays the network 
      itself by default.  [CTB]
    - Elementwise network operators now support multigraphs, and use 
      basic network attributes in a sane way.  Operator semantics have been
      substantially expanded for the muligraph case; hypergraphs are not
      yet supported, but missingness is.  [CTB, submitted by MSH]
    - Support for bipartite graphs in read.paj has been improved.  [MSH]
    - In addition to changes noted above, network.layout.fruchtermanreingold 
      uses a lightly different repulse radius by default; this seems to
      work better on large graphs.  [CTB]
    - plot.network has some overdue performance enhancements, including
      a more scalable Fruchterman-Reingold implementation.  Note that
      plot.network no longer coerces anything to adjacency form, although
      particular layout methods might.  [CTB]
  Bug Fixes:
    - add.edge and add.edges would crash when called with NA vertex IDs in the
      head or tail lists.  [CTB, submitted by Skye Bender-deMoll]
    - is.na.network failed when called on networks with deleted edges.  
      [CTB, submitted by MSH]
    - network.dyadcount handled NAs improperly in some cases.  [MSH]
    - network.initialize now stops with an error when called with <=0
      vertices, rather than producing undefined behavior.  [CTB, submitted
      by Skye Bender-deMoll]
    - Various minor issues in plot.network have bee fixed.  [CTB, mostly
      based on imports from gplot in sna; some gplot fixes contributed by Alex
      Montgomery]
    - print.network and print.summary.network could fail when called on
      multiplex or hypergraphic networks with the wrong matrix.type settings.
      [CTB]

v1.4-1 Minor Changes/Bug Fixes
  Changes:
    - print.summary.character now behaves in a more intuitive way (and always
      generates at least marginally useful output).  [CTB]
    - print.summary.network has been rewritten -- too much confusion about what
      it was supposed to do.  [CTB]
    - The na.omit option to print.summary.network and print.network should now
      be considered deprecated (it does essentially nothing at this point. [CTB]
  Bug Fixes:
    - print.summary.network was giving the same information twice (and other 
      information badly.  [CTB]
v1.4 New Functions, New Features, Changes, and Bug Fixes
  New Functions:
    - is.na.network: returns a network whose edges correspond to missing edges
      in the original network.  (Also supported by new backend function 
      isNANetwork_R.)  [CTB]
    - network.naedgecount: returns the number of missing edges in a network
      object.  [CTB]
    - Internal functions summary.character and print.summary.character have
      been added for use with network print/summary methods.  [MSH]
    - Internal function is.color has been added to allow heuristic
      identification of color names (for use with attribute plotting).  
      as.color similarly attempts to coerce its input into some reasonable
      color value for display purposes.  [CTB]
  New Features:
    - The network edge assignment operator ([<-.network) now allows NAs to be
      given as assignment values (resulting in missing edges if no attribute
      specified, or missing attribute values otherwise).  [CTB]
    - The C-level network API headers are now contained in the "inst" directory;
      from now on, they will be maintained there.  [CTB]
    - read.paj now imports vertex attributes.  [Patch submitted by Alexander
      Montgomery]
  Changes:
    - Many minor documentation updates (including adding references to the 
      recent JSS article).  [CTB]
    - CITATION file has been updated to reflect current R standards.  [CTB]
    - Color support in plot.network.default has been greatly expanded and
      rationalized.  [CTB]
    - is.adjacent now sets na.omit=FALSE by default; there seems to be a general
      consensus that this results in the more obvious pattern of behavior
      (i.e., missing edges from i to j result in a value of NA, unless there 
      are also non-missing (i,j) edges present).  The man page has also been
      updated to describe this behavior in greater detail.  [CTB]
    - Per the above, as.sociomatrix and related coercion methods also now
      display missing data information by default.  [CTB]
    - summary.network now returns a summary.network object, and printing takes
      place within print.summary.network (which is standard R behavior).
      Something approximately like this was being done before, but behavior
      should now be more conventional; summary.network objects can also carry
      optional information as network attributes with names of the form
      "summary.*". [CTB; Submitted by DRH]
    - plot.network.default now automagically displays labels if manually
      supplied (following the behavior of sna's gplot).  [CTB]
    - print.network now shows missing edge information.  [CTB]
    - The man page for network.dyadcount now emphasizes the fact that
      directed dyads are returned when is.directed(x)==TRUE.  (This was noted
      in the example, but was explicitly discussed in the main page.)  [CTB;
      Submitted by DRH]
    - as.matrix.network.edgelist and as.matrix.network.incidence now return
      degenerate matrices instead of NULL when called with an empty graph.
      [Pavel Krivitsky]
    - Undocumented support for the design matrix and special "respondent"
      attributes should be considered deprecated, and will be removed in the
      next version.  [CTB]
  Bug Fixes:
    - as.matrix.network.edgelist dropped the dimensions of its output when
      called on a graph with one edge.  [CTB]
    - as.matrix.network.incidence produced an error when called with an 
      empty graph, and generated incorrect matrices when called with graphs
      containing missing or previously deleted edges.  [CTB]
    - get.neighborhood was incorrectly including ego for loopless undirected
      graphs.  [CTB]
    - list.vertex.attributes produced an error when called on networks with
      different attributes on each vertex.  [CTB]
    - plot.network.default was not displaying colors as advertised in some
      cases [CTB; Submitted by Cori Mar]
    - print.network and summary.network were giving ugly output.  [MSH]
    - read.paj was reversing arc directions.  [CTB; Submitted by Kevin Lewis]
    - The initial startup message was giving the wrong help command.  [CTB;
      Submitted by Kevin Zembower]
    - read.paj was failing in some cases.  [Pavel Krivitsky]
    - summary.network didn't call the mixing matrix summary properly.  [MSH]
v1.3 New Functions, New Features, Changes, and Bug Fixes
  New Functions:
    - network.vertex.names<-: simplified assignment for vertex names.  [CTB]
  New Features:
    - A CITATION file has now been added to the package, to encourage good
      behavior.  The initial on-load announcements have been tweaked
      accordingly.  [CTB]
  Changes:
    - add.edges (via the backend routine add_edges_R) now adds edges in a
      more efficient way; substantial performance gains should be observed
      when adding multiple edges at one time, versus previous package
      versions.  [CTB]
    - network print and summary methods now consistently refer to "vertex 
      attributes" instead of "nodal attributes."  [CTB]
    - network constructors now set vertex names by default; this brings the
      actual behavior of the package in line with its apparent behavior (since
      network.vertex names and similar routines will "fake" vertex names if 
      none are present, thus producing results which are inconsistent with
      get.vertex.attribute in the latter case).  [CTB]
    - The DESCRIPTION file has been tweaked to be a bit more professional, and
      to add the statnet web site.  [CTB]
  Bug Fixes:
    - Fixed "virtual subsetting" for network objects when a single vector of
      virtual cell numbers is provided, or when a one-row, two-column matrix is
      given.  In both cases, network was treating the numbers as if they were
      first-column selectors.  [CTB]
    - Direct assignment to internal components of network objects was failing
      in certain cases. This has been fixed, although users should note that
      this behavior is both unsupported and generally a Bad Idea (TM).  Please
      use interface methods instead!  [CTB; Submitted by Skye Bender-deMoll]
    - Fixed bug in read.paj.  [MSH]
    - Fixed display of attribute information in summary.network.  [MSH]
    - plot.network is now compatible with the updated ergm package.  [CTB]
    - Previously implemented functionality in plot.network allowing one-word
      specification of vertex or edge attributes for display parameters was
      not working uniformly.  [CTB]
v1.2 New Functions, Changes, Bug Fixes
  New Functions:
    - $<-.network: replacement method for network objects.  [CTB]
    - sum.network, prod.network: sums and products of multiple network 
      objects.  [CTB]
  Changes:
    - Direct assignments with network objects on the right-hand side now copy
      their right-hand argument (per standard R semantics).  Originally, a 
      pointer to the right-hand argument was copied (and network.copy was 
      required for direct assignment).  The direct use of network.copy is now
      unnecessary for most purposes, and should be avoided.  [CTB]
    - network.density now allows explicit control of missing data behavior,
      support for ignoring "structural zeros" (per bipartite), and supports
      a wider range of hypergraphic cases.  [CTB]
    - Some adjustments have been made to the overloading of network arithmetic
      operators, to ensure compatibility with future versions of R.  Most
      importantly, the passing of an attrname argument to arithmetic operators
      is now defunct (since it violates the S3 generics).  The addition of
      more general sum and prod methods hopefully make up for this 
      regression.  [CTB]
    - The network extraction and assignment operators now behave more like 
      conventional matrices.  In particular, single vectors are assumed to 
      contain lists of cell indices (when given in isolation), and one-row,
      two-column matrices are treated as other two-column matrices.  [CTB]
    - Various minor documentation and test file updates.  [CTB]
  Bug Fixes:
    - as.matrix arguments have been modified to harmonize with the new R (2.5)
      generics definitions.  [CTB]
    - Annoying but harmless tracer messages have been removed.  [CTB]
    - Protection stack could overflow when large numbers of edges were deleted
      in a single call.  [CTB; Submitted by Pavel Krivitsky]
v1.1-2 Changes, Bug Fixes
  Changes:
    - getNeighborhood, getEdges, and getEdgeIDs (internal) now force
      type="combined" behavior on undirected networks; this was done at the R
      level before, but is now enforced in C as well.  This is not generally
      user-level transparent, but affects the experimental network API [CTB]
  Bug Fixes:
    - as.network.matrix was not setting the bipartite attribute of the returned
      network properly, when called with a non-FALSE bipartite argument [MSH]
    - An error was present in some error return functions, causing errors on
      errors (which, happily, were only relevant when an error occurred)
      [CTB; Submitted by Skye Bender-deMoll]
v1.1-1 New Functions, Changes
  New Functions:
    - The internal function setVertexAttribute has been added.  This has
      no immediate user-level effect, but the new function is supported
      in the C API [CTB]
  Changes:
    - Use of the protection stack has been changed, so as to avoid racking up
      huge stacks when creating very large networks.  This is expected to have
      a minimal impact on performance, but will avoid protection stack overflow
      issues in some cases  [CTB]
    - A change in R 2.4.0 has apparently made it impossible for generic 
      two-argument Ops (e.g., +,-,*) to dispatch to functions with more than
      two arguments.  A side effect of this is that "+.network" and friends
      must be called with the full function name (as opposed to simply "+") 
      when the optional attrname argument is being used.  Note that this is
      not a change in the network package (although the test code has been
      updated to reflect it), but a regression due to R.  Go complain to the
      R team  [CTB]
v1.1 New Features, Bug Fixes, Changes
  New Features:
    - [.network and [<-.network now allow the use of vertex names (where
      present) for selection of vertices [CTB]
  Bug Fixes:
    - add.vertices did not verify the integrity of vattr, and could generate
      a segfault if incorrectly called with a non-null, non-list value [CTB;
      reported by Skye Bender-deMoll]
    - as.network (and friends) could segfault if matrix.type was forced
      to adjacency while bipartite>0; new behavior essentially forces the
      use of the bipartite matrix method in this case [CTB]
    - delete.edges and set.edge.attribute returned an annoying (but harmless)
      warning when called with zero-length eid [CTB; reported by David Hunter]
    - delete.vertices did not adjust bipartite attribute (where present) to
      account for loss of mode 1 vertices [CTB]
    - get.vertex.attribute generated an error when called with na.omit=TRUE
      in some cases [CTB]
    - network.incidence could not be used to construct undirected dyadic
      networks [CTB]
    - set.vertex.attribute generated an error if called with attribute lists
      of length != network.size [CTB; reported by Skye Bender-deMoll]
  Changes:
    - Added a new overview man page (network-package) with information on how
      to get started with network [CTB]
    - [<-.network will now remove edges with zero values if both names.eval and
      add.edges are set, and will not add edges for those cells.  Previously,
      the standard behavior was to add edges for all cells [CTB]
    - Added delete.edges to the "see also" for add.edges [CTB; suggested by
      Skye Bender-deMoll]
    - permute.vertexIDs now throws a warning when called with a cross-mode
      vertex exchange on a bipartite graph [CTB]
    - Default matrix type for as.matrix.network is now "adjacency," rather than
      the output of which.matrix.type().  Coercion methods should not have
      variable behavior depending on features such as network size, even if
      it is convenient for some purposes!  The old behavior can be easily
      obtained via setting matrix.type=which.matrix.type(x), for those who
      want it [CTB]
v1.0-1 Minor Bug Fixes, Changes
  Bug Fixes:
    - Various warnings were removed (apparently, these only appeared in 
      R<2.1) [CTB]
    - plot.network was failing on networks where is.bipartite==TRUE [CTB]
  Changes:
    - The generic form of %c% was temporarily removed, to avoid namespace
      issues with sna.  (This will be rectified in future releases.) [CTB]
v1.0 New Functions, New Features, Changes, and Bug Fixes
  New Functions:
    - Operator overloading for +, -, *, |, &, and ! have been added, as has the
      composition operator, %c% [CTB]
    - Operator overloading is now supported for "[" and "[<-"; this allows
      network objects to be treated transparently as if they were adjacency
      matrices (in some cases, at least).  New extraction/replacement operators
      %n%, %n%<-, %v%, %v%<- have been added for extracting/assigning values to
      network and vertex attributes (respectively) [CTB]
    - network.copy: returns a copy of the input network [CTB]
    - network.dyadcount: return the number of dyads in a network (optionally
      adjusting for the missing dyads) [MSH]
  New Features:
    - add.edges now checks for illegal loop-like edges when edge.check==TRUE
      [CTB]
    - get.neighborhood now allows users to specify whether missing edges should
      be ignored [CTB]
    - set.edge.value now accepts edge values in vector format [CTB]
  Changes:
    - All access access functions now modify their arguments in place; this
      greatly improves performance, but may produce unexpected behavior.
      If users wish to generate a modified copy of a network, they must
      first generate the copy and then modify it.  Otherwise, the old
      object will be modified as well. In accordance with this, modification
      methods now return their (modified) arguments invisibly.  [CTB]
    - Most access functions have now been backended; this has improved the
      performance of many operations by as much as two orders of magnitude [CTB]
    - get.edges and get.edgeIDs now treat all undirected networks as if
      called with neighborhood=="combined" [CTB]
    - as.matrix.network.incidence now handles undirected edges in a more
      conventional way [CTB]
    - network.adjacency will now ignore diagonal entries if has.loops=FALSE 
      [CTB]
  Bug Fixes:
    - as.network.edgelist and as.network.incidence were producing spurious edge
      attributes [CTB]
    - list.edge.attributes generated failed under certain conditions (submitted
      by Matthew Wiener) [CTB]
    - set.edge.attribute was able to write attributes into non-existent (NULL)
      edges [CTB]
    - set.edge.value could exhibit strange behavior when carelessly chosen
      edge subsets were selected [CTB]
    
v0.5-4 New Facilities for Bipartite, New Features, Changes, and Bug Fixes
  New Functions:
    - network.bipartite to store an explicit bipartite network.
      Modified network.initialize, etc, to accept "bipartite" argument. [MSH]
    - is.bipartite: logical test for a bipartite network [MSH]
    - read.paj: read one or more network objects from a Pajek file [MSH, DS]
  New Features:
    - summary.network now reports on edge attributes [MSH]
  Changes:
    - The composition operator (%c%) has been removed due to a name conflict
      with the sna package; since sna now supports network objects, its
      version can be used instead. [CTB]
    - as.sociomatrix is now properly configured to work in tandem with 
      as.sociomatrix.sna (in the sna package).  The functionality of the 
      routine has also been extended slightly.  [CTB]
  Bug Fixes:
    - .First.lib: Print out correct welcome banner for package [MSH]
    - Fix displayisolates determination in plot.network.default [MSH]

v0.5-3 New Functions, New Data Set, and Changes
  New Functions:
    - permute.vertexIDs: Permute vertices in their internal representation [CTB]
  New Data:
    - emon: Drabek et al.'s Emergent Multi-organizational Networks [CTB]
  Changes:
    - The obsolete examples directory has been removed. [CTB]

v0.5-2 New Features, New Functions and Bug Fixes
  New Functions:
    - delete.vertices: Remove one or more vertices (and associated edges)
      from a network object.
    - delete.edge.attribute, delete.network.attribute, delete.vertex attribute:
      Remove an edge/network/vertex attribute.
    - list.edge.attributes, list.network.attributes, list.vertex attributes:
      List all edge/network/vertex attribute names.
  New Features:
    - plot.graph.default now accepts vertex/edge attribute names for most
      vertex/edge display properties.
  Bug Fixes:
    - Edge deletion produced exciting and unexpected behavior in some cases.
    - network.initialize set vertex na attributes to TRUE by default.