mirror of
https://github.com/Mercury-Language/mercury.git
synced 2025-12-15 05:44:58 +00:00
a9208b47a8
Previously, we specialized calls to string.format and io.format only if
the format string specified no flags, widths, precisions or non-default
bases for any conversion. We now specialize calls to those predicates
regardless of the format string. To make this possible, we now have two
copies of the code to parse format strings.
library/string.parse_runtime.m:
This new module, carved out out library/string.format.m, contains
one of those copies. As the name suggests, this copy is used by
the runtime system.
compiler/parse_string_format.m:
This new module contains the other copy. As its location suggests,
this copy is used by the compiler. The separate copy is necessary
because the compiler's parsing needs are different from the runtime's.
The main reason for the separation is that the compiler needs to be
able to handle cases where widths and predicates are given by manifest
constants, as well as when they are given by the values of variables
at runtime (when the format string contains conversion specifiers
such as "%*d"). Having the runtime handle this extra complexity
would have inevitably increased the runtime's format string interpretation
overhead, which is undesirable.
The other reason is that compiler can afford to postprocess the result
of the parsing in order to avoid having to concatenate two or more
constant strings at runtime.
library/string.parse_util.m:
Types and predicates that are common to the library and compiler copies
of the format string parsing code.
compiler/format_call.m:
We used to check whether format strings match the supplied list
of values to be printed by invoking string.format on dummy inputs
of the supplied types, and seeing whether you got an exception.
We now call parse_string_format.m instead. Not only does this
avoid shenanigans with exceptions (which the deep profiler cannot
yet handle), but it also allows us to specialize all calls
to string.format and io.format. We therefore do so.
We used to specialize calls to io.format by creating a single string
to be printed (using the machinery needed to specialize calls to
string.format), and then printing that. This did unnecessary memory
allocations to hold the intermediate strings. We now simply print
each component one after the other, which avoids this overhead.
library/string.format.m:
Add versions of the predicates that print chars, strings, ints and floats
that are specialized to each possible situation about whether the caller
specifies a width and/or a precision. This means that format_call.m
doesn't have to generate code that allocates cells on the heap.
Remove the code that was moved to new submodules of string.m.
compiler/simplify_proc.m:
List the predicates in string.format.m that the compiler
may generate calls to in the list of such predicates that we maintain
to prevent dead_pred_elim from deleting them before format_call.m
gets a chance to do its work.
compiler/module_imports.m:
If the code of the module calls a predicate named "format" that
may refer to string.format or io.format, then implicitly import
the modules that contain the types and predicates that the code
generated by format_call.m will refer to. Note that since this
module_imports.m works on Mercury code that has not been module
qualified yet, we have to use a conservative approximation.
Reorganize the way we discover what library modules we have to
implicitly import. Instead of a separate boolean for each piece
of functionality that needs an implicit import, put them together
into a structure. Due to Peter's work on argument packing some
years back, this is now more efficient as well as more convenient
that passing around a bunch of booleans. Also switch to passing
around the structure as an accumulator, instead of having to do
bool.ors all over the place.
Since the code for computing the list of modules to be implicitly
imported can sometimes add a library module to the list twice,
sort that list and remove any duplicates.
To make this possible, remove the version of the main predicate
that appends the implicitly imported module names to a list of
module names passed by our caller, since we don't want to sort
THAT list of module names.
Give the remaining version of that predicate that does this
a name that better matches the names of related predicates.
compiler/modules.m:
Conform to the change in module_imports.m. Mark some suspicious code
with XXXs.
Remove some redundant comments.
compiler/options.m:
doc/user_guide.texi:
Add a new option that tells format_call.m what it should do
when it finds more than one problem with a format string: whether
it should print a message only for the first problem, or for all
the problems. (The default is the latter.)
library/MODULES_DOC:
A list of the all the Mercury source files in the library that
should be included in the user guide.
library/MODULES_UNDOC:
A list of the all the Mercury source files in the library that
should NOT be included in the user guide.
library/library.m:
Include the new publicly visible new submodule of string.m,
(string.parse_util). And both string.parse_util and string.parse_runtime
(which is NOT visible from outside string.m) to the list of modules
in the Mercury standard library.
Update the rules for where new library modules should be documented
to account for non-publically-visible submodules, which we haven't had
before. Document the need to include new modules in either MODULES_DOC
or MODULES_UNDOC.
mdbcomp/builtin_modules.m:
Provide convenient access for format_call.m to the names of the
submodules of string.m that it needs access to.
library/Mmakefile:
Add a new target, check_doc_undoc, which checks whether any source files
are missing from MODULES_DOC or MODULES_UNDOC.
doc/Mmakefile:
Read the list of modules to be documented from MODULES_DOC. This replaces
old code that tried to list all the undocumented modules, but missed one:
it referred to erlang_conf.m instead of erlang_builtin.m.
When creating the library documentation, filter out any lines that
contain the string NOTE_TO_IMPLEMENTORS. We now use this mechanism
to include reminders to implementors in what would otherwise be a
publically visible part of string.m.
tools/bootcheck:
Copy MODULES_DOC and MODULES_UNDOC to the stage 2 and 3 library
directories, since the check_doc_undoc target, which is invoked by
"make all" in the library directory, needs them.
compiler/hlds_out_module.m:
When dumping out the table of types, dump out discriminated union types
in a style that follows our recommended coding style.
20 lines
322 B
Plaintext
20 lines
322 B
Plaintext
backjump.m
|
|
erlang_builtin.m
|
|
erlang_rtti_implementation.m
|
|
mer_std.m
|
|
mutvar.m
|
|
par_builtin.m
|
|
private_builtin.m
|
|
profiling_builtin.m
|
|
region_builtin.m
|
|
robdd.m
|
|
rtti_implementation.m
|
|
stm_builtin.m
|
|
string.format.m
|
|
string.parse_runtime.m
|
|
string.parse_util.m
|
|
string.to_string.m
|
|
table_builtin.m
|
|
term_size_prof_builtin.m
|
|
test_bitset.m
|