Mercury Bugs - mercury
View Issue Details
IDProjectCategoryView StatusDate SubmittedLast Update
0000564mercuryBugpublic2022-09-08 14:272022-09-08 15:35
Reporterwangp 
Assigned To 
PrioritynormalSeverityminorReproducibilityalways
StatusnewResolutionopen 
PlatformOSOS Version
Product Version 
Target VersionFixed in Version 
Summary0000564: unnecessary use_module in .opt files
DescriptionI was trying to figure out why a seemingly small change to one module causes a project-wide recompilation when developing with intermodule module optimisation enabled.


For the attached test case, mmc --make-opt-int fib.m produces:

:- module fib.
:- use_module builtin.
:- use_module int.
:- use_module io.
:- use_module map.
:- use_module private_builtin.

:- func fib.fib(int) = int.
:- mode fib((builtin.in)) = (builtin.out) is det.

:- pragma inline(func((fib.checked_fib)/1)).
fib.checked_fib(V_3) = V_4 :-
    ( if
      V_5 = 40 : int,
      int.(V_3 >= V_5)
    then
      V_4 = -1 : int
    else
      V_4 = fib.fib(V_3)
    ).

I don't see the need for:

:- use_module io.
:- use_module map.

The io module is only used by a predicate not exported or opt-exported from fib.m.
The map module is not used at all.

Not sure about private_builtin either.
TagsNo tags attached.
Relationships
Attached Files? fib.m (551) 2022-09-08 14:27
https://bugs.mercurylang.org/file_download.php?file_id=323&type=bug

Notes
(0001203)
zs   
2022-09-08 15:35   
Unfortunately, this is not so much a bug, as an as-yet-unimplemented feature,
because at the moment, we just generate a use_module declaration for
*every* module imported by the module whose .opt file we are creating,
*regardless* of whether it is needed or not. See the XXXs in intermod.m
starting at line 1386.

I think the reason for this non-attempt was that originally, .trans_opt
files used to be written out piecemeal; each analysis pass, after it was run,
appended its results to that file. This made it complicated to find out
up front which modules would be needed, and writing out the use_modules
at the end could have screwed up the then-rigid ordering requirements
of the make_hlds pass. Such considerations would have affected .opt files
as well, because .opt files are also written out in two pieces by two
separate algorithms, as explained by the comment starting on line 58.
However, the above is speculation, since I am not intermod.m's original
author.

As for why builtin and private_builtin are in the .opt file:
they are *always* implicitly imported and used respectively in every module,
whether needed or not, and intermod.m blindly copies them to the .opt file.

Issue History
Date ModifiedUsernameFieldChange
2022-09-08 14:27wangpNew Issue
2022-09-08 14:27wangpFile Added: fib.m
2022-09-08 15:35zsNote Added: 0001203