mirror of
https://github.com/Mercury-Language/mercury.git
synced 2026-04-21 04:13:46 +00:00
20b77c995e
compiler/parse_dcg_goal.m:
The DCG transformation adds unifications to the then-parts and else-parts
of if-then-elses. Use the contexts of the actual user code in these parts
for the compiler-generated conjunctions of which they are part.
tests/invalid/dcg_context.{m,err_exp}:
The test case from m-users which motivated this change. The error message
generated by the compiler for this code used the same context for
both the user-written then-part and the compiler-created else part
of an if-then (no else) goal in a DCG clause. The fix makes it clear,
in this instance and most others, that when mode analysis complains
about a mismatch between the two arms of the if-then-else, it is
talking about two separate pieces of code.
tests/invalid/Mmakefile:
Enable the new test case.
81 lines
3.3 KiB
Mathematica
81 lines
3.3 KiB
Mathematica
%---------------------------------------------------------------------------%
|
|
% vim: ts=4 sw=4 et ft=mercury
|
|
%---------------------------------------------------------------------------%
|
|
%
|
|
% This test is derived from a piece of code posted to m-users on 2023 oct 28
|
|
% that the compiler generated error messages for, but for which those messages
|
|
% were hard to understand:
|
|
%
|
|
% dcg_context.m:076: The variable `Class' is ground in some branches but not
|
|
% dcg_context.m:076: others.
|
|
% dcg_context.m:076: In this branch, `Class' is free.
|
|
% dcg_context.m:076: In this branch, `Class' is ground.
|
|
% dcg_context.m:076: The variable `MaybeSuper' is ground in some branches but
|
|
% dcg_context.m:076: not others.
|
|
% dcg_context.m:076: In this branch, `MaybeSuper' is free.
|
|
% dcg_context.m:076: In this branch, `MaybeSuper' is ground.
|
|
%
|
|
% The cause of the bug was the absence of an else part in the nested if.
|
|
% Since DCG clauses may have if-thens without elses, this is not itself a bug,
|
|
% but it would have been the job of this else part to bind Class and
|
|
% MaybeSuper, and mode analysis reported the absence of producers for
|
|
% these two variables in the compiler-generated else part of that if-then.
|
|
%
|
|
% The compiler generated the misleading contexts because it used the
|
|
% context of the if-then as a whole, which is the context of the "if" token,
|
|
% as the context of both
|
|
%
|
|
% - the compiler-generated then part, which consists of the actual code
|
|
% of the then-part in the source code, plus unifications of DCG vars, and
|
|
% - the compiler-generated else part, which consists of just unifications
|
|
% of DCG vars.
|
|
%
|
|
% We have no better context to use for the else part, but for the then part,
|
|
% we can use the context of the then part in the source code. In most programs,
|
|
% this should be different from the context of the "if" token, and the
|
|
% error message will be easi*er* to understand. It is still not easy in that
|
|
% the context we print for the else part will be at a *smaller* line number
|
|
% than the context for the then part, but the part of parse_dcg_goal.m
|
|
% that parses such goals has no access to any context that would be more
|
|
% meaningful, and such a context may not even exist. (Just consider code
|
|
% in which one line contains the whole then part and the parentheses closing
|
|
% the if-then, and following line containing the code following the if-then
|
|
% as a whole.)
|
|
%
|
|
% Note that this fix does not help if the then-part is on the same line
|
|
% as the "if" token. However, in that case, the programmer can blame
|
|
% only himself/herself for the lack of helpfulness of the line numbers
|
|
% in the error message :-(
|
|
%
|
|
%---------------------------------------------------------------------------%
|
|
|
|
:- module dcg_context.
|
|
:- interface.
|
|
|
|
:- import_module list.
|
|
:- import_module maybe.
|
|
|
|
:- type ps
|
|
---> ps(int, string).
|
|
|
|
:- type sr_stack == list(sr).
|
|
|
|
:- type sr
|
|
---> tk(int, string)
|
|
; sexp(float, list(sr)).
|
|
|
|
:- pred get_class_super(ps::out, maybe(ps)::out,
|
|
sr_stack::in, sr_stack::out) is det.
|
|
|
|
:- implementation.
|
|
|
|
get_class_super(Class, MaybeSuper) -->
|
|
( if [tk(Cp, Cb)] then
|
|
{ Class = ps(Cp, Cb) },
|
|
{ MaybeSuper = no }
|
|
else if [sexp(_, [tk(Cp, Cb), tk(Sp, Sb)])] then
|
|
{ Class = ps(Cp, Cb) },
|
|
{ MaybeSuper = yes(ps(Sp, Sb)) }
|
|
% The bug is that there is no else case here.
|
|
).
|