1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
|
## The strength and weakness of predicate options {#predopts-pro-cons}
Many ISO predicates accept options, e.g., open/4, write_term/3. Options
offer an attractive alternative to proliferation into many predicates
and using high-arity predicates. Properly defined and used, they also
form a mechanism for extending the API of both system and application
predicates without breaking portability. I.e., previously fixed
behaviour can be replaced by dynamic behaviour controlled by an option
where the default is the previously defined fixed behaviour. The
alternative to using options is to add an additional argument and
maintain the previous definition. While a series of predicates with
increasing arity is adequate for a small number of additional
parameters, the untyped positional argument handling of Prolog quickly
makes this unmanageable.
The ISO standard uses the extensibility offered by options by allowing
implementations to extend the set of accepted options. While options
form a perfect solution to maintain backward portability in a linear
development model, it is not well equipped to deal with concurrent
branches because
1. There is no API to find which options are supported in a
particular implementation.
2. While the portability problem caused by a missing predicate in
Prolog _A_ can easily be solved by implementing this predicate, it
is much harder to add processing of an additional option to an
already existing predicate.
Different Prolog implementations can be seen as concurrent development
branches of the Prolog language. Different sets of supported options
pose a serious portability issue. Using an option _O_ that establishes
the desired behaviour on system _A_ leads (on most systems) to an error
or system _B_. Porting may require several actions:
- Drop _O_ (if the option is not vital, such as the layout options
to write_term/3)
- Replace _O_ by _O2_ (i.e., a differently named option doing the
same)
- Something else (cannot be ported; requires a totally
different approach, etc.)
Predicates that process options are particularly a problem when writing
a compatibility layer to run programs developed for System _A_ on System
_B_ because complete emulation is often hard, may cause a serious
slowdown and is often not needed because the application-to-be-ported
only uses options that are shared by all target Prolog implementations.
Unfortunately, the consequences of a partial emulation cannot be
assessed by tools.
## Options as arguments or environment? {#predopts-environment}
We distinguish two views on options. One is to see them as additional
parameters that require strict existence, type and domain-checking and
the other is to consider them `locally scoped environment variables'.
Most systems adopt the first option. SWI-Prolog adopts the second: it
silently ignores options that are not supported but does type and domain
checking of option-values. The `environment' view is commonly used in
applications to create predicates supporting more options using the
skeleton below. This way of programming requires that _pred1_ and
_pred2_ do not interpret the same option differently. In cases where
this is not true, the options must be distributed by _some_pred_. We
have been using this programming style for many years and in practice it
turns out that the need for active distribution of options is rare.
I.e., options either have distinct names or multiple predicates
implement the same option but this has the desired effect. An example of
the latter is the =encoding= option, which typically needs to be applied
consistently.
==
some_pred(..., Options) :-
pred1(..., Options),
pred2(..., Options).
==
As stated before, options provide a readable alternative to high-arity
predicates and offer a robust mechanism to evolve the API, but at the
cost of some runtime overhead and weaker consistency checking, both at
compiletime and runtime. From our experience, the `environment' approach
is productive, but the consequence is that mistyped options are silently
ignored. The option infrastructure described in this section tries to
remedy these problems.
## Improving on the current situation {#predopts-improving}
Whether we see options as arguments or locally scoped environment
variables, the most obvious way to improve on the current situation is
to provide reflective support for options: discover that an argument is
an option-list and find what options are supported. Reflective access to
options can be used by the compiler and development environment as well
as by the runtime system to warn or throw errors.
### Options as types {#predopts-as-types}
An obvious approach to deal with options is to define the different
possible option values as a type and type the argument that processes
the option as list(<option_type>), as illustrated below. Considering
options as types fully covers the case where we consider options as
additional parameters.
==
:- type open_option ---> type(stream_type) |
alias(atom) | ... .
:- pred open(source_sink, open_mode, stream, list(open_option)).
==
There are three reasons for considering a different approach:
- There is no consensus about types in the Prolog world, neither about
what types should look like, nor whether or not they are desirable. It
is not likely that this debate will be resolved shortly.
- Considering options as types does not support the `environment'
view, which we consider the most productive.
- Even when using types, we need reflective access to what
options are provided in order to be able to write
compile or runtime conditional code.
### Reflective access to options {#predopts-reflextion}
From the above, we conclude that we require reflective access to find
out whether an option is supported and valid for a particular predicate.
Possible option values must be described by types. Due to lack of a type
system, we use library(error) to describe allowed option values.
Predicate options are declared using predicate_options/3:
* [[predicate_options/3]]
* [[assert_predicate_options/4]]
The predicates below realise the support for compile and runtime
checking for supported options.
* [[current_predicate_option/3]]
* [[check_predicate_option/3]]
The predicates below can be used in a development environment to inform
the user about supported options. PceEmacs uses this for colouring
option names and values.
* [[current_option_arg/2]]
* [[current_predicate_options/3]]
The library can execute a complete check of your program using
check_predicate_options/0:
* [[check_predicate_options/0]]
The library offers predicates that may be used to create declarations
for your application. These predicates are designed to cooperate with
the module system.
* [[derive_predicate_options/0]]
* [[retractall_predicate_options/0]]
* [[derived_predicate_options/3]]
* [[derived_predicate_options/1]]
|
