mirror of
https://github.com/Mercury-Language/mercury.git
synced 2026-04-24 13:53:54 +00:00
083d376e6598628362ee91c2da170febd83590f4
4 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Peter Wang
|
d69bb27dfb |
Avoid require_tailrec_N failures at higher optimisation levels.
tests/invalid/require_tailrec_1.m:
tests/invalid/require_tailrec_2.m:
tests/invalid/require_tailrec_3.m:
Suppress some inlining that occurs at higher optimisation levels,
that changes the warnings produced for require_tail_recursion
pragmas.
tests/invalid/require_tailrec_1.err_exp:
tests/invalid/require_tailrec_1.err_exp2:
tests/invalid/require_tailrec_1.err_exp3:
tests/invalid/require_tailrec_2.err_exp:
tests/invalid/require_tailrec_2.err_exp2:
tests/invalid/require_tailrec_2.err_exp3:
Update expected outputs for changed line numbers.
|
||
|
Zoltan Somogyi
|
eb4d94d7a0 |
Make require_tailrec_N pass with deep profiling.
tests/invalid/require_tailrec_1.err_exp3:
tests/invalid/require_tailrec_2.err_exp3:
tests/invalid/require_tailrec_3.err_exp3:
Add these expected output files for deep profiling grades. They
differ from the expected output files for both the LLDS and MLDS
backends because deep profiling does not support *any* tail calls.
tests/invalid/require_tailrec_1.m:
tests/invalid/require_tailrec_2.m:
tests/invalid/require_tailrec_3.m:
Update the source code of these test cases to document
what circumstances each expected output file is for.
tests/invalid/require_tailrec_1.err_exp:
tests/invalid/require_tailrec_1.err_exp2:
tests/invalid/require_tailrec_2.err_exp:
tests/invalid/require_tailrec_2.err_exp2:
tests/invalid/require_tailrec_3.err_exp:
tests/invalid/require_tailrec_3.err_exp2:
Update the existing expected output files to account for the
changes in line numbers caused by the new documentation.
|
||
|
Zoltan Somogyi
|
3a1ed2efcb |
Replace simple_call_id with pf_sym_name_arity.
compiler/prog_data.m:
Delete the simple_call_id data type, since it is isomorphic
to the pf_sym_name_arity type, which more clearly specifies
what is stored inside it.
compiler/prog_out.m:
Rename (all three versions of) simple_call_id_to_string to
pf_sym_name_orig_arity_to_string, both to conform to the change
in the input data type, and to emphasize that the resulting string
will contain the *original* arity of functions (which does not include
the return value), which is one less than the arity in the
pf_sym_name_arity structure (which, in accordance with the
convention inside the compiler that the arity is the length
of the argument list, *does* include the return value).
Delete the provision inside simple_call_id_to_string, now
pf_sym_name_orig_arity_to_string, for special handling of the names
of the predicates we use to implement promises, because it seems that
*none* of the call sites to any of the three versions of this function
can actually pass to it the identity of such a predicate. These calls
refer to a predicate or mode declaration item (which promise predicates
do not have), to clause or foreign_proc items (which again, promise
predicates cannot have) or calls to the predicate (promise predicates
cannot be called).
Delete the exported predicate simple_call_id_to_sym_name_arity.
It was called from exactly one place, inside prog_out.m itself,
and this diff inlines that call.
Avoid unnecessary forwarding of work from prog_out.m to error_util.m.
Delete (all three versions of) write_simple_call_id. The changes
below replace all their (few) uses with calls to
pf_sym_name_orig_arity_to_string.
compiler/error_util.m:
Replace the simple_call() error piece with qual_pf_sym_name_orig_arity,
and add a new version unqual_pf_sym_name_orig_arity. Their names
explicitly say that they print the original arities of functions,
also say whether they strip away any module qualification on the
sym_name inside the pf_sym_name_arity. We prefer the unqual version
in situations where the module qualifier is implicit, which usually means
that it must be the same as the name of the module being compiled, because
it reduces visual clutter for readers of error messages.
Put {qual,unqual}_pf_sym_name_orig_arity next to their most closely
related function symbols, {qual,unqual}_sym_name_arity, in the
format_component type. (This yields a few inconsequential changes
in the order of error_specs when sorted.)
compiler/add_class.m:
compiler/add_clause.m:
compiler/add_foreign_proc.m:
compiler/add_pragma.m:
compiler/add_pragma_tabling.m:
compiler/add_pragma_type_spec.m:
compiler/add_pred.m:
compiler/check_libgrades.m:
compiler/check_parse_tree_type_defns.m:
compiler/check_promise.m:
compiler/check_typeclass.m:
compiler/convert_parse_tree.m:
compiler/equiv_type.m:
compiler/goal_expr_to_goal.m:
compiler/hlds_data.m:
compiler/hlds_desc.m:
compiler/hlds_goal.m:
compiler/hlds_out_pred.m:
compiler/hlds_out_util.m:
compiler/hlds_pred.m:
compiler/inst_util.m:
compiler/llds_out_instr.m:
compiler/make.module_target.m:
compiler/make.program_target.m:
compiler/make_hlds_error.m:
compiler/make_hlds_warn.m:
compiler/mark_tail_calls.m:
compiler/ml_unify_gen_construct.m:
compiler/mlds_to_c_stmt.m:
compiler/mode_errors.m:
compiler/mode_info.m:
compiler/modecheck_util.m:
compiler/module_qual.collect_mq_info.m:
compiler/module_qual.qualify_items.m:
compiler/parse_item.m:
compiler/parse_module.m:
compiler/parse_type_repn.m:
compiler/polymorphism.m:
compiler/post_typecheck.m:
compiler/pre_quantification.m:
compiler/pred_table.m:
compiler/prog_item.m:
compiler/recompilation.version.m:
compiler/term_constr_build.m:
compiler/termination.m:
compiler/type_ctor_info.m:
compiler/typecheck.m:
compiler/typecheck_errors.m:
compiler/typecheck_info.m:
compiler/unify_gen_construct.m:
compiler/unused_imports.m:
Conform to the changes above. Where I am pretty sure that in an error
message, the module qualifier of a name must be the current module,
use unqual_pf_sym_name_orig_arity instead of qual_pf_sym_name_orig_arity.
Add some sanity checks where they seem appropriate.
Replace sequences of io.write_strings with uses of io.format where this
yields clearer code. (This is why there were no remaining calls to
write_simple_call_id.)
Shorten some too-long lines.
In add_pred.m, make the order of some predicate definitions match
the order of the calls to them.
tests/invalid/ambiguous_method.err_exp:
tests/invalid/ambiguous_method_2.err_exp:
tests/invalid/bad_pred_arity.err_exp:
tests/invalid/bad_sv_unify_msg.err_exp:
tests/invalid/bigtest.err_exp:
tests/invalid/bug113.err_exp:
tests/invalid/bug197.err_exp:
tests/invalid/bug278.err_exp:
tests/invalid/bug410.err_exp:
tests/invalid/bug476.err_exp:
tests/invalid/bug487.err_exp:
tests/invalid/complex_constraint_err.err_exp:
tests/invalid/constrained_poly_insts.err_exp:
tests/invalid/errors.err_exp:
tests/invalid/errors1.err_exp:
tests/invalid/errors2.err_exp:
tests/invalid/exported_mode.err_exp:
tests/invalid/field_syntax_error.err_exp:
tests/invalid/foreign_singleton.err_exp:
tests/invalid/funcs_as_preds.err_exp:
tests/invalid/imported_mode.err_exp:
tests/invalid/invalid_binary_literal.err_exp:
tests/invalid/invalid_float_literal.err_exp:
tests/invalid/invalid_hex_literal.err_exp:
tests/invalid/invalid_octal_literal.err_exp:
tests/invalid/make_opt_error.err_exp:
tests/invalid/missing_det_decls.err_exp:
tests/invalid/multimode_syntax.err_exp:
tests/invalid/null_char.err_exp:
tests/invalid/occurs.err_exp:
tests/invalid/record_syntax_errors.err_exp:
tests/invalid/ref_to_implicit_pred.err_exp:
tests/invalid/require_tailrec_1.err_exp:
tests/invalid/require_tailrec_1.err_exp2:
tests/invalid/require_tailrec_2.err_exp:
tests/invalid/require_tailrec_2.err_exp2:
tests/invalid/require_tailrec_3.err_exp:
tests/invalid/require_tailrec_3.err_exp2:
tests/invalid/state_vars_test2.err_exp:
tests/invalid/state_vars_test3.err_exp:
tests/invalid/state_vars_test5.err_exp:
tests/invalid/type_inf_loop.err_exp:
tests/invalid/typeclass_constraint_extra_var.err_exp:
tests/invalid/typeclass_mode_2.err_exp:
tests/invalid/typeclass_test_12.err_exp:
tests/invalid/typeclass_test_2.err_exp:
tests/invalid/typeclass_test_9.err_exp:
tests/invalid/types.err_exp:
tests/invalid/types2.err_exp:
tests/invalid/unbound_type_vars.err_exp:
tests/invalid/with_type.err_exp:
tests/invalid_purity/purity_nonsense.err_exp:
tests/invalid_purity/purity_nonsense2.err_exp:
tests/warnings/double_underscore.exp:
tests/warnings/pragma_source_file.exp:
tests/warnings/singleton_test.exp:
tests/warnings/singleton_test.exp2:
tests/warnings/singleton_test.exp3:
tests/warnings/singleton_test.exp4:
tests/warnings/singleton_test_state_var.exp:
tests/warnings/warn_return.exp:
tests/warnings/warn_return.exp2:
tests/warnings/warn_return.exp3:
tests/warnings/warn_succ_ind.exp:
tests/warnings/warn_succ_ind.exp2:
tests/warnings/warn_succ_ind.exp3:
tests/warnings/warn_succ_ind.exp4:
Update all these expected output files. Most changes are due to
predicate and function names no longer being module qualified when
the module qualification is obvious. A few changes are due to the
change in the relative ordering of the function symbols of the
format_component type.
For singleton_test, warn_return and warn_succ_ind in warnings,
the changes to the non-C .exp files were done by hand, so they
may have the wrong white space.
|
||
|
Zoltan Somogyi
|
756ed140d3 |
Improve non-tail recursion warnings from the MLDS code generator.
Consider two mutually recursive procedures p and q, where p tailcalls q
but q does not tailcall p. The LLDS backend can optimize the tailcall
from p to q, but the MLDS backend cannot, because it can optimize only
*mutual* tail recursion. In such cases, and in other cases where the
MLDS cannot implement a recursive call as a tail call even though the
LLDS backend can, add to the warning message we generate a description
of the reason why the MLDS code generator cannot do what the programmer
may expect them to do.
compiler/ml_gen_info.m:
Include in the state of the MLDS code generator tail-call-related
information not just about all the procedures in the current TSCC,
but about all the procedures in the current SCC. And instead of
storing warning messages about non-tail recursive calls in the code
generator state, store a list of the non-tail recursive calls themselves.
These two changes together allow us to gather a description of all
the non-tail recursive calls before we generate even the first warning.
This should allow a later to use this information to generate messages
such as "the recursive call from p to q here cannot be optimized
because the recursive call from q to p at this *other* context
cannot be optimized".
Make the names of some types and function symbols clearer.
compiler/ml_call_gen.m:
Whenever a recursive call cannot be optimized as a tail call,
record the reason behind that fact.
Use pred_proc_ids instead of separate pred_ids and proc_ids,
since this simplifies the code and should speed paramater passing.
compiler/ml_proc_gen.m:
Pass tail-call-related information about the procedures of the SCC along
as we generate code for the SCC's procedures.
Once we have generated code for all of the SCC's procedures,
turn the final list of the non-tail calls in the SCC into a list
of warnings for them.
compiler/options.m:
doc/user_guide.texi:
Add a compiler option that controls whether we report as non-tail
calls recursive calls that *obviously* cannot be tail calls, because
they are followed by other recursive calls. The standard example
is the first recursive call in qsort.
A comment by Paul has long called for such an option, and this diff
touches the same code anyway.
Fix an unrelated old blemish: document both of the old options
--warn-non-tail-recursion-{self,mutual}, but in a commented out form,
since they are internal-use options. Before this diff, only one was
documented, and only user_guide.texi.
compiler/mark_tail_calls.m:
Add parameters to the predicates that generate warnings that specify
why the recursive call cannot be optimized as a tail call (describing
the reason if it is a result of a limitation in the compiler, and
not a problem in the program being compiled), and whether the call
is obviously a non-tail call.
When marking a call as recursive, but the recursive call is obviously
not a tail call, mark it as such using a new goal feature. The MLDS
backend needs this to allow it to filter warnings for obviousness
the same way as mark_tail_calls.m itself.
compiler/hlds_goal.m:
Add the new goal feature mentioned above.
compiler/ml_closure_gen.m:
compiler/ml_tailcall.m:
compiler/rtti_to_mlds.m:
compiler/saved_vars.m:
Conform to the changes above.
tests/invalid/require_tailrec_1.err_exp2:
tests/invalid/require_tailrec_2.err_exp2:
tests/invalid/require_tailrec_3.err_exp2:
Add the expected warnings from the MLDS code generator.
tests/invalid/Mercury.options:
Specify -E for one of the tests above, to test the handling of the
verbose-only part of the warnings.
tests/EXPECT_FAIL_TESTS.hlc.gc:
Do not expect tests/invalid/require_tailrec_[123] to fail
in grade hlc.gc anymore.
|