Estimated hours taken: 10
Allow modules to be put in source files whose names do not directly match
their the module names. When looking for the source for a module such
as `foo:bar:baz', search for it first in `foo.bar.baz.m', then in `bar.baz.m',
and finally in `baz.m'.
compiler/prog_io.m:
Change prog_io__read_module so that it returns the name of
the module read, as determined by the `:- module' declaration.
Add predicate `check_module_has_expected_name', for checking
that this name matches what was expected.
compiler/modules.m:
Add read_mod_from_file, for reading a module given the file name,
and generated_file_dependencies, for generating the dependencies
of a module given the file name. (As opposed to the module name.)
Change read_mod and read_mod_ignore_errors so that they
search for `.m' files as described above, and return the name
of the source file read.
Also improve the efficiency of read_dependencies slightly:
when reading in `.int' files, there's no need to call
split_into_submodules, because we generate a seperate
`.int' file for each submodule anyway.
compiler/mercury_compile.m:
Change the handling of command-line arguments.
Arguments ending with `.m' are assumed to be file names,
and other arguments are assumed to be module names.
For file names, call read_mod_from_file instead of read_mod.
compiler/handle_options.m:
Change help message to reflect the above change to the semantics
of command-line arguments.
compiler/intermod.m:
compiler/trans_opt.m:
Fix a bug: call prog_io__read_opt_file instead of prog_io__read_module.
doc/user_guide.texi:
Document the above change to the semantics of command-line arguments.
Update the "libraries" chapter to reflect our support for nested
modules.
tests/*/*.m:
tests/*/*.exp:
Fix a few incorrect module names in `:- module' declarations.
Estimated hours taken: 0.5
Avoid about half of the slow "mmake realclean"s required by a bootcheck.
tests/*/runtests:
Concentrate all the actions performed before the test and after
a successful tests (both of which involve an "mmake realclean")
into two scripts, tests/{startup,shutdown}.
tests/shutdown:
Clean up the directory, and touch the file CLEAN.
tests/startup:
If the file CLEAN exists and is the most recent file in the directory,
consider the directory clean to beging with. Otherwise, run mmake
realclean.
where switch detection was getting confused by explicit existential
quantifications.
compiler/switch_detection.m:
When detecting switches, traverse through some/2 goals.
Also simplify the code a bit.
tests/valid/Mmakefile:
tests/valid/some_switch.m:
Regression test for the above change.
where switch detection was getting confused by explicit existential
quantifications.
compiler/switch_detection.m:
When detecting switches, traverse through some/2 goals.
Also simplify the code a bit.
tests/valid/Mmakefile:
tests/valid/some_switch.m:
Regression test for the above change.
Estimated hours taken: 0.2
tests/runtests:
tests/*/runtests:
When reporting the results, print the options that controlled
the tests, so you don't have to search for them.
tests/*/runtests:
Clean up the directory if the tests are successful, in order
to reduce disk space usage.
Estimated hours taken: 0.5
Makefile:
tests/valid/Mmakefile:
tests/invalid/Mmakefile:
Fix some code rot in the rules for `make clean';
amoung other things, change them so that they
handle `--use-subdirs' correctly.
Estimated hours taken: 0.25
tests/valid/runtests:
tests/valid/Mmakefile:
Change a hard-coded test for `$file.o' so that it will
work with --use-subdirs.
Estimated hours taken: 6
Fix some problems with the `--use-subdirs' option.
The compiler itself now compiles and bootstraps fine with --use-subdirs.
compiler/modules.m:
Put `.h' files in the source directory, rather than
in the `Mercury/hs' subdirectory.
compiler/mercury_compile.m:
scripts/Mmake.rules:
If `--use-subdirs' is enabled, pass `-I.' to the C compiler,
so that #include statements work relative to the source directory
rather than relative to the `Mercury/cs' subdirectory.
scripts/Mmake.vars.in:
Define $(cs_subdir), $(os_subdir) etc. variables;
these are set to the directory to use for .c, .o, etc. files,
(including the trailing `/'), or to the empty string,
if --use-subdirs is not set.
scripts/Mmake.rules:
Use $(cs_subdir), $(os_subdir) etc. to avoid the code
duplication created by my previous change to handle
--use-subdirs.
Also add lots of comments, and reorder the code
in a more logical order.
Mmakefile:
library/Mmakefile:
compiler/Mmakefile:
profiler/Mmakefile:
tests/term/Mmakefile:
tests/valid/Mmakefile:
Use $(cs_subdir), $(os_subdir) etc. to fix a few hard-coded
file-names (e.g. *.dep, *_init.[co], tree234.o) that were
used in some of the rules.
library/Mmakefile:
Add `rm -f tags' to the rule for `mmake realclean'.
tools/bootcheck:
Add `--use-subdirs' option (defaults to setting of the
MMAKE_USE_SUBDIRS environment variable).
Change the code which compares the stage2 & stage3 C files
to use the appropriate location for them based on the
setting of this option.
Estimated hours taken: 5
Fix a bug in the optimization where polymorphism.m passes a
base_type_info in place of a type_info for non-polymorphic types.
The type of the variable was `base_type_info' not `type_info'.
This caused inlining.m to be unable to compute a type substitution,
and code_util__cons_id_to_tag aborted on an unsubstituted type
variable.
compiler/mercury_builtin.m
compiler/code_util.m
Add a new builtin, unsafe_type_cast/2, used by common.m
to preserve type correctness.
Make unsafe_promise_unique/2 a builtin, since it is basically
the same as unsafe_type_cast/2.
compiler/polymorphism.m
Set the type of a `base_type_info' passed where a `type_info'
is expected to `type_info'.
Don't add the type_info argument for unsafe_type_cast, since it is
not needed and would make the code in common.m more complicated.
compiler/common.m
Generate a call to unsafe_type_cast rather than an assignment
unification when the assignment would not be type correct.
tests/valid/inlining_bug.m
Regression test.
tests/general/common_type_cast.m
tests/general/common_type_cast.exp
Test of type casts.
Estimated hours taken: 0.5
Fix a bug reported by Tomas By which caused multiple
definition errors when two modules used a third
module, and one of the modules imported the other.
compiler/modules.m
Delay processing of short interfaces until all long
interfaces have been read to avoid reading both
the long and short interface for a module.
tests/valid/module_*.m
Add a test case.
Estimated hours taken: 0.1
tests/valid/intermod_lambda2.m
tests/valid/intermod_test2.m
Remove some mode errors which were not detected
because mode analysis was never run - only `.opt'
files are produced for these modules.
Estimated hours taken: 0.25
tests/valid/agc_graph.m:
Modify the code to avoid a (spurious in this case, as it happens)
"Sorry, not implemented" message for abstract exported polymorphic
equivalence types whose bodies have fewer type variables than their
heads.
Estimated hours taken: 1
compiler/quantification.m:
Fix a problem where the use of explicit quantifiers lead to
spurious warnings about unbound type variables, by deleting
explicit quantifiers during quantification (once the
variables have been renamed apart, the explicit quantifiers
are no longer needed).
tests/valid/Mmakefile:
tests/valid/explicit_quant.m:
Regression test.
Estimated hours taken: 1
compiler/modules.m:
Fix a bug introduced in stayl's changes to add support for
`use_module': it was not reading in the `.int2' files
for modules indirectly imported from `.int' files.
tests/valid/Mmakefile:
tests/valid/module_a.m:
tests/valid/module_b.m:
tests/valid/module_c.m:
tests/valid/module_d.m:
A regression test for the above-mentioned bug.
Estimated hours taken: 1
Add new options for disabling some warnings, so that they
can be disabled by `--inhibit-warnings'.
compiler/options.m:
Add new options `warn-missing-module-name'
and `warn-wrong-module-name' (enabled by default),
and ensure that `--inhibit-warnings' disables these options.
compiler/prog_io.m:
Add code to conditionalize the printing of warnings
based on the settings of the new options.
doc/user_guide.texi:
Document the new options.
tests/valid/Mmakefile:
tests/valid/inhibit_warn_test.m:
A regression test.
Estimated hours taken: 25
Handle inference of mostly-unique (not just unique) modes.
compiler/unique_modes.m:
Change the handling of calls so that if the mode originally selected
by ordinary mode analysis (modes.m) won't work, then it now calls
modecheck_call_pred to introduce a new mode for the predicate.
compiler/unify_proc.m:
Generalize the existing "unify request queue" so that
it can handle requests for any kind of predicate, not
just for unification predicates, and hence rename the
`unify_request' table which contains it as the `proc_request' table.
compiler/hlds_module.m:
Rename the `unify_request' table as the `proc_request' table.
compiler/modecheck_call.m:
When introducing new modes for predicates, do this by adding them
to the request queue. This ensures that when we do get around
to processing predicates added in unique mode analysis, we will
run mode analysis, switch detection, and determinism analysis
before doing unique mode analysis. To make this work, we
need to pass down a `maybe(determinism)' argument indicating
the determinism of the newly added procedure. unique_modes.m
sets this to the determinism inferred for the mode originally
inferred by modes.m. That should be OK, because uniqueness
should not affect determinism.
Also change modecheck_call_proc to return a `Changed' flag
indicating whether we need to rerun fixpoint analysis.
And change `get_var_insts_and_lives' to return `dead' for
mostly_unique modes, not just for unique modes, so that
we can infer `mdi' modes.
compiler/modes.m:
Instead of analyzing everything to a fixpoint, and then
analyzing the queued unification procedures, change it so
that analysis of queued procedures is part of each fixpoint
iteration.
When doing ordinary mode analysis for calls, pass `no' as the
new `maybe(determinism)' argument to modecheck_call_pred.
compiler/constraint.m:
compiler/cse_detection.m:
Ignore the new output argument for modecheck_call_proc.
compiler/mode_util.m:
Add code to insts_to_mode to recognize `muo', `mui', and `mdi'
modes and use the corresponding abbreviations (previously this
was done only for `uo', `ui', `di', `in' and `out').
tests/valid/mostly_uniq_mode_inf.m:
A test case for the above change.
Estimated hours taken: 8
Clean up the handling of unbound type variables.
Fix a bug with unbound type variables in lambda expressions.
Run purity analysis, modechecking etc. even if there were type errors.
compiler/mercury_compile.m:
Run purity analysis, modechecking etc. even if there were type
errors. This fixes a bug (inconsistency between the code and
the comments) that seems to have been introduced in stayl's
change to mercury_compile.m (revision 1.25) to add intermodule
unused argument elimination: the comment said "continue,
even if type checking found errors", but the code did not
continue.
This change was needed to ensure that we still report a warning
message about unused type variables for tests/invalid/error2.m;
without it, we stop after type checking and don't do purity
analysis, and so don't report the warning.
compiler/typecheck.m:
compiler/purity.m:
Move the code for checking for unbound type variables
from typecheck.m to purity.m. It needs to be done
*after* type inference has been completed, so it
can't be done in the ordinary type checking/inference
passes. Add code to purity.m to bind the
unbound type variables to the builtin type `void'.
compiler/polymorphism.m:
Comment out old code to bind unbound type variables
to `void'; the old code was incomplete, and this
is now done in purity.m.
compiler/notes/compiler_design.html:
Document the above changes.
tests/valid/Mmakefile:
tests/valid/unbound_tvar_in_lambda.m:
Regression test for the above-mentioned bug with unbound type
variables in lambda expressions.
tests/warnings/singleton_test.exp:
tests/invalid/errors2.err_exp:
Change the expected warning message for unbound type variables.
The error context is not as precise as it used to be, I'm afraid:
we only know which function/predicate the error occurred in,
not which clause. Also it now comes out in a different order
relative to the other error messages.
tests/invalid/errors2.err_exp:
tests/invalid/funcs_as_preds.err_exp:
Add some new error/warning messages that are output now that
we run mode and determinism analysis even if there are type errors.
Estimated hours taken: 4
Change unique mode analysis (unique_modes.m) so that it does the same
fixpoint analysis that ordinary mode analysis (modes.m) does.
This is a first step towards mode inference of "mostly-unique" modes.
Unique mode analysis still won't introduce any new modes, but it may
now change the final insts of inferred modes from `unique' to `mostly_unique'.
The next step (not included in this diff) will be to change the handling of
calls in unique_modes.m so that it does introduce new modes.
Also change things so that the inferred modes are normally printed
only after unique mode analysis, with determinism annotations attached.
compiler/modes.m:
Add an extra WhatToCheck parameter to lots of predicates,
so they can also be used for checking unique modes.
Add a few small new bits of code to handle the
`WhatToCheck = check_unique_modes' case.
compiler/unique_modes.m:
Call the routines in modes.m rather than special casing things.
The checking of goals is still special-cased, since for
example unique_modes does not do reordering, but the traversal
of modules and procedures now uses code from modes.m.
compiler/mode_errors.m:
Add an extra bool parameter to write_mode_inference_messages
indicating whether or not to print determinism annotations.
tests/valid/Mmakefile:
tests/valid/mostly_uniq_mode_inf.m:
Test case for the above change.
Estimated hours taken: 0.5
tests/Mmake.common:
tests/valid/Mmakefile:
For the tests of accurate gc, just add `--gc accurate'
to GRADEFLAGS, rather than setting GRADE=asm_fast.agc.
Otherwise the tests fail on systems for which we
don't support asm_fast.* grades.
Estimated hours taken: 1
Fix a bug in the writing of .opt files.
The problem occurs when you have a pragma inline declaration for an
exported function. The declaration:
:- func addone(int) = int.
:- pragma inline(addone/1).
addone(I) = I + 1.
gets written to the .opt file as:
:- pragma inline((foo:addone)/2).
That is, the arity of the _predicate_ version is written rather than
the arity of the _function_.
compiler/intermod.m:
compiler/mercury_to_mercury.m:
Add a pred_or_func argument to mercury_output_pragma_decl,
and use that to determine the declared arity.
tests/valid/intermod_test.m:
tests/valid/intermod_test2.m:
Regression test.
Estimated hours taken: 4
compiler/liveness.m:
Recompute nonlocal sets before starting liveness computations
in order to avoid a compiler abort. The new version could in
some cases also lead to the generation of better code through
saving fewer variables across disjuncts.
tests/valid/liveness_nonlocals.m:
Test case to exercise the fix.
tests/valid/Mmakefile:
Enable the new test case.
with common sub-expression; mercury 0.7 failed this test, reporting
"Software Error: modecheck fails when repeated", due to confusion
between h.o. _function_ call and h.o. _predicate_ call.
compiler/hlds_goal.m:
Add `pred_or_func' field to HLDS higher_order_calls.
compiler/modes.m:
compiler/modecheck_call.m:
compiler/hlds_out.m:
compiler/*.m:
Add code to handle new field for higher_order_call goals.
tests/valid/Mmake:
tests/valid/ho_func_call.m:
Regression test for the above-mentioned bug.
Estimated hours taken: 8
Better handling of abstract unification of insts which always fail.
compiler/det_analysis.m:
New predicate det_par_conjunction_detism/3 to compute the
determinism of parallel conjunctions (or virtual parallel
conjunctions).
compiler/hlds_data.m:
Make the ground_inst_table store the determinism of the
grounding operation. The reason for this is that
make_ground_inst* now returns the determinism of the
grounding operation, and so it makes sense to store the
determinisms along with cached groundings to avoid
recomputing them.
compiler/inst_util.m:
Add a new argument to return the determinism of the
unification/grounding to the following predicates:
abstractly_unify_inst_functor
abstractly_unify_bound_inst_list
abstractly_unify_bound_inst_list_lives
abstractly_unify_inst_list_lives
make_ground_inst_list_lives
make_ground_inst_list
make_ground_inst
make_ground_bound_inst_list
This is to make it easier to locate unifications which
cannot succeed.
compiler/instmap.m:
If you attempt to insert a not_reached into an instmap, the
entire instmap becomes unreachable.
compiler/modecheck_unify.m:
Slight reorganisation of code dealing with var-functor
unifications. Simple optimisations have been moved to
modecheck_unify_functor.
All unifications which definitely fail are now optimised
to `fail'.
compiler/mode_util.m:
Minor change to support the alteration to the ground_inst_table
above.
tests/valid/Mmake:
tests/valid/empty_bound_inst_list.m:
Test case.
Estimated hours taken: 0.25
tests/valid/Mmake:
Change the rule for `mmake clean' to remove *.err, not *.err2,
since the `*.err2' files have been renamed as `*.err'.
Estimated hours taken: 0.25
tests/valid/Mmake:
Add proper dependencies so that `mmake depend' does the right thing.
Previously it failed for the recently-added multi-module test
`func_int_bug_main.m'.
Estimated hours taken: 2
compiler/simplify.m:
Fix a bug where the code generator complained about `nondet
model in semidet context': If a model_det or model_semi
if-then-else has a model_nondet condition, then change the
determinism of the if-then-else as a whole to nondet and put it
inside a `some'. (This can only happen if the `then' part
has determinism `erroneous' or `failure'.) The treatment
of this problem for if-then-elses is similar to the way we
treat the analgous problem for conjunctions.
tests/valid/Mmake:
tests/valid/tricky_ite.m:
Regression test for the above bug fix.
Estimated hours taken: 2
compiler/mercury_to_mercury.m:
Fix a bug or two that showed up when exporting a func/0 with a func
declaration that combines types and modes. It was incorrectly
(a) printing the determinism in the type declaration, not just
the mode declaration and (b) printing the determinism in the
wrong place in the mode declaration.
tests/valid/Mmake.m:
tests/valid/func_int_bug.m:
tests/valid/func_int_bug_main.m:
Regression test for the above bug fix.
Estimated hours taken: 4
compiler/frameopt:
Bug fix: make sure that the transformation that keeps stack frames
put the instruction that restores succip before a livevals/goto pair.
It used to put it in between the livevals and the goto, which violates
an invariant that value numbering depends on.
compiler/vn_table:
Allow the two operands of an operator to be the same.
compiler/vn_util:
Add simplification rules for dealing with boolean and and or,
and rules exploiting the identities of several operators.
tests/hard_coded/float_map:
A test case for the fix to builtin_float_compare recently checked in
by Fergus.
tests/hard_coded/Mmake:
Enable the new test case.
tests/valid/livevals_seq:
A new test case for the fix to frameopt.
tests/valid/double_vn:
A new test case for the fix to vn_table.
tests/valid/Mmake:
Enable the new test cases.
compiler/simplify.m:
Fix a bug: the code was passing around two different copies
of ModuleInfo, one passed directly, one stored in the det_info
in the simplify_info, and these were not being kept in sync.
That caused the final ModuleInfo to contain the wrong merge_inst
table, which led to map__lookup failures.
tests/valid/Mmake:
tests/valid/lazy_list.m:
Add regression test for the above bug.
for variables in lambda goals.
compiler/quantification.m:
(1) When implicitly-quantifying a lambda goal, the "lambda-outsidevars"
should be set to the empty set, so that variables occurring
only inside disjoint lambda goals are locally quantified
inside those lambda goals.
(2) When quantifying conjunctions and if-then-elses, use both
the lambda outsidevars and the ordinary outsidevars when
computing the non-locals, rather than just using the
ordinary outsidevars.
tests/valid/lambda_quant.m:
Add another test case for bug (1).
There was already a tests in this module that was supposed to
test this sort of stuff, but that test case happened to work,
because the effects of bugs (1) and (2) cancelled out. :-(
tests/valid/Mmake:
tests/valid/lambda_quant_bug.m:
Add a regression test for bug (2).
Estimated hours taken: 5
Fix several bugs in liveness for accurate GC.
compiler/liveness.m:
- Add `liveness__get_nonlocals_and_typeinfos'.
- Call `liveness__get_nonlocals_and_typeinfos' instead of just
getting nonlocals. This removes a lot of repeated code (and
improves correctness).
- Intersect the initially live variables with the non-locals and
their typeinfo vars, not just the non-locals.
- Don't add corresponding typeinfos to the liveness after the
intersection (previously we did this, which "fixed" many
problems that doing the intersection wrong introduced, but
introduced a few bugs of its own).
tests/valid/Mmake:
tests/valid/agc_graph.m:
tests/valid/agc_ho_pred.m:
Test cases for these bugs.
Estimated hours taken: 0.25
tests/valid/Mmake:
tests/valid/higher_order_implied_mode.m:
Regression test for bug in modecheck_call.m with handling
of implied modes of higher-order calls.
Estimated hours taken: 0.5
compiler/simplify.m
Fix a bug in simplify.m where deconstructions were being
propagated into lambda expressions, changing the non-locals of
the lambda expression, but the uni-modes field of the lambda
expression construction was not being updated.
tests/valid/lambda_struct_bug.m
Regression test.
Estimated hours taken: 1
Fix a bug in liveness that causes a sanity check to fail for accurate
gc.
compiler/liveness.m:
Handle if-then-else correctly for .agc grades, it wasn't
handling polymorphic type variables correctly.
tests/valid/Mmake:
tests/valid/agc_ite.m:
Add a test case for if-then-else - previously the compiler
aborted on this test case.
Estimated hours taken: 0.1
tests/valid/Mmake:
tests/valid/ho_inst.m:
Test case to ensure that special_preds for higher order
func insts are created and mode/determinism checked correctly.
Estimated hours taken: 5
Fix another bug triggered by compilation for accurate gc.
The compiler was not correctly handling the unbound type variables when
doing inlining.
compiler/inlining.m:
Update TypeVarSet and TypeInfoVarMap when inlining - unbound
type variables can occur in the callee which need to be migrated
to the caller when inlined. Introduce new inline_info type, which
stores information threaded through inling. This should make future
changes to inlining much easier.
compiler/type_util.m:
Add `apply_substitutions_to_var_map' which will update a
map(tvar, var) given a type substitution and a variable
substitution.
tests/valid/Mmake:
tests/valid/agc_unbound_typevars2.m:
Test case for this bug.
Estimated hours taken: 0.25
tests/valid/Mmake:
tests/valid/pred_with_no_modes.m:
Add a regression test for the bug in polymorphism.m that I just fixed.
Estimated hours taken: 0.1
compiler/intermod.m
Use the declared argmodes or procedures when gathering modes to put
into .opt files, since these are the modes that are written to the
.opt file.
tests/valid/intermod_lambda.m
tests/valid/intermod_lambda2.m
Updated these to act as a regression test.
Estimated hours taken: 5
Remove support for term_to_type and type_to_term implemented as special
preds. Remove support for one-cell and one-or-two-cell type_infos (now
shared-one-or-two-cell type_infos). Move definitions that were in
mercury_builtin.m back to where they belong.
This code has been removed because it is no longer used, and was no
longer being maintained but was still quite complex.
tests/general/disj_disj.m:
tests/general/dnf.m:
tests/general/higher_order.m:
tests/general/nondet_disj.m:
tests/hard_coded/cc_nondet_disj.m:
tests/hard_coded/pragma_inline.m:
tests/invalid/funcs_as_preds.err_exp:
tests/misc_tests/mdemangle_test.exp:
tests/valid/agc_unbound_typevars.m:
tests/valid/middle_rec_labels.m:
tests/valid/subtype_switch.m:
tests/warnings/infinite_recursion.m:
Import module `list' or `term' (or both).
Estimated hours taken: 1
Add test cases for recently fixed accurate garbage collection compilation
problems.
tests/valid/Mmake:
tests/valid/agc_unused_in.m:
Test case for getting liveness right with an unused polymorphic typed
variable with a mode of `in'.
tests/valid/agc_unbound_typevars.m:
Test case for keeping the information needed for unbound type
variables.
Estimated hours taken: 0.25
tests/valid/Mmake:
tests/valid/switch_detection_bug.m:
Add a regression test for the bug in switch detection
that I just fixed.
Estimated hours taken: 1.5
Fix a determinism analysis abort which was caused by det_analysis.m not
updating the instmap to include the initial insts of the lambda variables
before processing a lambda goal. This problem was also present in
cse_detection.m and simplify.m.
tests/valid/Mmake
tests/valid/lambda_instmap_bug.m
Regression test.
Estimated hours taken: 0.5
tests/valid/Mmake:
tests/valid/type_inf_ambig_test.m:
Add test case to check that type inference checks for
ambiguities predicate-at-a-time rather than clause-at-a-time,
as per my recent change to typecheck.m.
