Mercury Bugs - mercury
View Issue Details
IDProjectCategoryView StatusDate SubmittedLast Update
0000142mercuryBugpublic2010-04-06 15:072010-07-15 15:47
Reporterwangp 
Assigned Towangp 
PrioritynormalSeverityminorReproducibilityalways
StatusresolvedResolutionfixed 
PlatformOSOS Version
Product Version 
Target VersionFixed in Version 
Summary0000142: problem updating rtti varmaps on inlining
DescriptionIn the following test case, higher order specialisation creates
`or__ho1' from of `or'.

We try to inline `or__ho1' into `orr'. In the call to `or__ho1' the
same variable `TypeClassInfo_for_dcg_14' appears twice in the argument
list, corresponding to the formal arguments V_22 and
TypeClassInfo_for_dcg_18 in the callee.

Eventually we try to rename the variables in the callee rtti varmap.
We hit an assertion failure in apply_substs_to_constraint_map
(hlds_rtti.m:777) as the two callee variables holding typeclass_infos
map to the same variable in the caller, and supposedly it is for two
different constraints. They are the same, just with different type
variables.

Is the fix to relax the assertion, or something else?

Additional Information% mmc --inline-single-use --optimise-higher-order dcg
Uncaught Mercury exception:
Software Error: hlds_rtti.m: Unexpected: inconsistent typeclass_infos
Stack dump follows:
   0 pred exception.throw/1-0 (erroneous) (exception.m:411)
   1 pred require.error/1-0 (erroneous) (require.m:122)
   2 pred libs.compiler_util.unexpected/2-0 (erroneous) (compiler_util.m:88)
   3 pred hlds.hlds_rtti.apply_substs_to_constraint_map/7-0 (det) (hlds_rtti.m:804)
   4 pred tree234.foldl/4-0 (det) (tree234.m:2625)
   5 pred map.foldl/4-0 (det) (map.m:962)
   6 pred hlds.hlds_rtti.apply_substitutions_to_rtti_varmaps/5-0 (det) (hlds_rtti.m:679)
   7 pred transform_hlds.inlining.do_inline_call/13-0 (det) (inlining.m:774)
   8 pred transform_hlds.inlining.inlining_in_call/11-0 (det) (inlining.m:652)
   9 pred transform_hlds.inlining.inlining_in_goal/4-0 (det) (inlining.m:558)
  10 pred transform_hlds.inlining.inlining_in_conj/4-0 (det) (inlining.m:860)
  11 pred transform_hlds.inlining.inlining_in_goal/4-0 (det) (inlining.m:571)
  12 2* pred transform_hlds.inlining.inlining_in_conj/4-0 (det) (inlining.m:860 and others)
  14 pred transform_hlds.inlining.inlining_in_goal/4-0 (det) (inlining.m:571)
  15 pred transform_hlds.inlining.inline_in_proc/5-0 (det) (inlining.m:489)
  16 3* pred transform_hlds.inlining.do_inlining/6-0 (det) (inlining.m:260 and others)
  19 pred transform_hlds.inlining.inlining/2-0 (det) (inlining.m:249)
  20 pred top_level.mercury_compile_middle_passes.maybe_do_inlining/6-0 (det) (mercury_compile_middle_passes.m:941)
  21 pred top_level.mercury_compile_middle_passes.middle_pass/7-0 (det) (mercury_compile_middle_passes.m:189)
  22 pred top_level.mercury_compile.mercury_compile_after_front_end/11-0 (det) (mercury_compile.m:1650)
  23 pred top_level.mercury_compile.mercury_compile/11-0 (det) (mercury_compile.m:1582)
  24 pred top_level.mercury_compile.compile/13-0 (det) (mercury_compile.m:1507)
  25 pred list.map_foldl2/7-2 (det) (list.m:2359)
  26 pred top_level.mercury_compile.compile_all_submodules/13-0 (det) (mercury_compile.m:1296)
  27 pred top_level.mercury_compile.process_module_2/8-0 (det) (mercury_compile.m:1269)
  28 pred top_level.mercury_compile.process_module/8-0 (det) (mercury_compile.m:1186)
  29 pred top_level.mercury_compile.process_arg_2/8-0 (det) (mercury_compile.m:882)
  30 pred top_level.mercury_compile.process_arg_build/10-0 (det) (mercury_compile.m:844)
  31 pred make.util.build_with_module_options_args_invoked/12-0 (det) (make.util.m:987)
  32 pred make.util.build_with_module_options_args/11-0 (det) (make.util.m:940)
  33 pred top_level.mercury_compile.process_arg/8-0 (det) (mercury_compile.m:817)
  34 pred top_level.mercury_compile.process_arg_list/10-0 (det) (mercury_compile.m:788)
  35 pred top_level.mercury_compile.process_args/8-0 (det) (mercury_compile.m:740)
  36 pred top_level.mercury_compile.process_all_args/8-0 (det) (mercury_compile.m:546)
  37 pred top_level.mercury_compile.main_after_setup/7-0 (det) (mercury_compile.m:422)
  38 pred top_level.mercury_compile.real_main_after_expansion/3-0 (det) (mercury_compile.m:308)
  39 pred top_level.mercury_compile.real_main/2-0 (det) (mercury_compile.m:118)
  40 pred top_level.main/2-0 (det) (top_level.m:56)
TagsNo tags attached.
Relationships
Attached Files? dcg.m (1,474) 2010-04-06 15:07
https://bugs.mercurylang.org/file_download.php?file_id=99&type=bug

Notes
(0000263)
wangp   
2010-04-21 11:54   
I think the problem may be this. inlining.get_type_substitution, which is used in a few places, creates a substitution from the callee's head argument types to types in the caller. However, it doesn't account for type variables in the range of functional dependencies, so they never appear in the type substitution.

So two separate type variables in the calleee remain as two different type variables after inlining, but they should have been unified. Which is why we got the assertion failure above: a single variable is supposedly holding the TypeClassInfo for two "different" constraints, but the two constraints only differ because their type variables weren't unified.

Does that make sense?
(0000277)
wangp   
2010-07-15 15:47   
committed a workaround 0000012:0000010 May

Issue History
Date ModifiedUsernameFieldChange
2010-04-06 15:07wangpNew Issue
2010-04-06 15:07wangpFile Added: dcg.m
2010-04-21 11:54wangpNote Added: 0000263
2010-07-15 15:47wangpNote Added: 0000277
2010-07-15 15:47wangpStatusnew => resolved
2010-07-15 15:47wangpResolutionopen => fixed
2010-07-15 15:47wangpAssigned To => wangp