Branch: refs/heads/davem/xs_refactor6
Home: https://github.com/Perl/perl5
Commit: 9a13ff8192343ee39ff106c4978ac47c8c4aaa56
https://github.com/Perl/perl5/commit/9a13ff8192343ee39ff106c4978ac47c8c4aaa56
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Constants.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/CountLines.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Eval.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Node.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Utilities.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/Typemaps.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/Typemaps/Cmd.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/Typemaps/InputMap.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/Typemaps/OutputMap.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/Typemaps/Type.pm
M dist/ExtUtils-ParseXS/lib/perlxs.pod
Log Message:
-----------
ParseXS: bump version 3.55 => 3.56
Commit: d4997b059bc22463dbadfe8a42ca2fe1ef5f69b2
https://github.com/Perl/perl5/commit/d4997b059bc22463dbadfe8a42ca2fe1ef5f69b2
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
Log Message:
-----------
ParseXS: fix code comment typo
Commit: dc0d8fb6b09cb8b006fc07e3adefb5d9e842a218
https://github.com/Perl/perl5/commit/dc0d8fb6b09cb8b006fc07e3adefb5d9e842a218
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
Log Message:
-----------
ParseXS: fix generate_output(): s/next/return/
A recent piece of refactoring by me, v5.41.4-92-g9c24c4ed29, moved
some error-checking code from OUTPUT_handler() to generate_output().
The former is in a loop, the latter isn't. So in
if (bad) { blurt(..); next }
the 'next' needs changing to 'return'.
It turns out no tests yet exercise this error condition, so it it wasn't
spotted at the time.
Commit: 8e55c50bd94dcb81a2d3d11da6cbf4d55946b79e
https://github.com/Perl/perl5/commit/8e55c50bd94dcb81a2d3d11da6cbf4d55946b79e
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Node.pm
Log Message:
-----------
ParseXS: refactor: use a var rather than $1
During earlier refactoring, the captures of a particular match were
assigned to variables; but a few lines further down, the code was still
using '$1' rather than the variable which $1 had been assigned to.
This commit makes it use the var instead.
No functional changes.
Commit: 2f114a88673d49fb91d3c8cc48410755e02bf9aa
https://github.com/Perl/perl5/commit/2f114a88673d49fb91d3c8cc48410755e02bf9aa
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: add tests for RETVAL TARG optimisation
There is an optimisation whereby the OP_ENTERSUB's TARG is sometimes
used to return the result, rather than creating a new mortal.
This optimisation has been present for many years, but was untested.
This commit adds tests, including ones for custom typemap entries
of the sort that can still be optimised.
It also tests for the optimisation in conjunction with OUTLIST args.
It also adds tests for when the output code is specified explicitly
on the OUTPUT line rather than being retrieved from a typemap. Currently
this isn't actually optimised.
Commit: c0541adfaa9627f6dbfa435d0b159526da5b93a8
https://github.com/Perl/perl5/commit/c0541adfaa9627f6dbfa435d0b159526da5b93a8
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
Log Message:
-----------
ParseXS: refactor: move RETVAL TARG optimise code
The perl code which does the emitting of the C code for the RETVAL value
return using TARG if possible, is in a different place than the general
RETVAL code emitting. This commit moves them into the same function.
For now this is a crude cut and paste: any simplifying of the resultant
function will be done later. For now, it's good just that more C code
generation functionality is concentrated in one place rather than being
diffused around all parts of ParseXS.pm.
In more detail: before this commit, the code looked like:
sub process_file {
...
if (RETVAL needs returning) {
if (have template code from OUTPUT line) {
# just directly emit the template code
}
elsif (can do TARG optimisation) {
# directly emit code to do PUSHi() etc
}
else {
$self->generate_output({ var => 'RETVAL', ...});
}
}
...
}
sub generate_output {
...
if ($var eq 'RETVAL') {
# emit generic C code to return RETVAL to the caller
}
...
}
After this commit, it looks like:
sub process_file {
...
if (RETVAL needs returning) {
$self->generate_output({ var => 'RETVAL', ...});
}
...
}
sub generate_output {
...
if ($var eq 'RETVAL') {
if (have template code from OUTPUT line) {
# just directly emit the template code
return;
if (can do TARG optimisation) {
# directly emit code to do PUSHi() etc
}
else {
# emit generic C code to return RETVAL to the caller
}
}
...
}
Commit: 0e9d0908cff755bc5c7dd1a3a463fe72a9104741
https://github.com/Perl/perl5/commit/0e9d0908cff755bc5c7dd1a3a463fe72a9104741
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: add more tests for OUTPUT keyword
In particular, test that the 'SETMAGIC: EN/DISABLE' modifiers work,
and that duplicate vars are reported.
Commit: 742941d53162eeebd51e899c3e285d0874198871
https://github.com/Perl/perl5/commit/742941d53162eeebd51e899c3e285d0874198871
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
Log Message:
-----------
ParseXS: refactor: simplify dup OUTPUT RETVAL
If RETVAL appears twice in the OUTPUT section, it will be reported
as an error (using blurt()), but parsing will continue.
During parsing of the second RETVAL, it is treated as an ordinary
parameter rather than special-cased as RETVAL. This doesn't seem useful,
although ultimately it doesn't matter, as the parse will be reported as
failing at the end anyway, so it doesn't matter what code gets
generated in the meantime.
This commit makes RETVAL still be treated as RETVAL on second appearance
in OUTPUT, which get makes the code slightly simpler.
Also, change the dup-detecting lines above it from a postfix 'if' to a
block 'if', so that the flow-control 'next' appears prominently on its
own line.
Commit: e0dc4e93d83c3171c1da67c2a19fd1fdd094928d
https://github.com/Perl/perl5/commit/e0dc4e93d83c3171c1da67c2a19fd1fdd094928d
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: add tests: 'RETVAL not in OUTPUT' warning
If the parser sees RETVAL in a CODE section, but there isn't an OUTPUT
section for the RETVAL, then it emits a warning. Test for this warning.
Note that it currently warns only if the OUTPUT section is not present;
it doesn't warn if the OUTPUT is present but doesn't contain RETVAL.
The next commit will fix this.
Commit: 894c447f99f3eefa55efcf3e562f4f39182ec96b
https://github.com/Perl/perl5/commit/894c447f99f3eefa55efcf3e562f4f39182ec96b
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: make RETVAL in CODE not OUTPUT warn more
ParseXS currently emits a warning if RETVAL is seen in a CODE section
but there is no OUTPUT section. This commit extends this so it warns even
if there *is* an OUTPUT section but that section doesn't contain
RETVAL.
To stop one of our existing tests giving a false positive warning, this
commit now makes it ignore 'PERL_UNUSED_VAR(RETVAL)' in a CODE section
in terms of 'have seen RETVAL in CODE', since this isn't a real use of
the variable.
Note that the original intent of adding this warning was that, without an
OUTPUT
RETVAL
the emitted code would still do XSRETURN(1), but nothing would be stored
at ST(0), so whatever was returned on the stack at that slot would be
whatever was already at that position on the stack, which might be one
of the passed args, or random garbage. This is true regardless of
whether OUTPUT is completely absent, or just RETVAL missing from OUTPUT.
So it's better to warn in both cases, not just the former.
Doing this also means that the 'xsub_seen_OUTPUT' field member is no longer
needed, simplifying the code.
Commit: 6ff7391379878297d55da3400200fb0cc2641c9d
https://github.com/Perl/perl5/commit/6ff7391379878297d55da3400200fb0cc2641c9d
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: fix, test 'OUTPUT foo not a param' err
If a var is listed in an OUTPUT section, but is not actually a
parameter, then the parser outputs an error message:
Error: OUTPUT foo not a parameter
This commit:
- Fixes a typo in the error message: it was previously "not an parameter".
- Extends the warning to include RETVAL where the XSUB is void.
Previously, it would plough on, and then give an error that it
couldn't find 'void' in the typemap, which is less obvious what the
problem is.
- Adds tests - this warning previously had no tests for it.
Commit: f64c07d524ae6418f28ab53bf0de788b3af16b8a
https://github.com/Perl/perl5/commit/f64c07d524ae6418f28ab53bf0de788b3af16b8a
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Node.pm
Log Message:
-----------
ParseXS: refactor: make OUTPUT update param fields
Currently, an E::PXS::Node::Param object each created for each parameter
declared in the XSUB's signature. This object is then updated with any
extra data that appears in an INPUT line.
This commit extends that so that the Param object is also updated with
any extra data from an OUTPUT line.
This allows us to remove the $self->{xsub_map_varname_to_seen_in_OUTPUT}
field.
The next few commits will build upon this to simplify various bits of
code, now that the Param object contains OUTPUT state too.
Commit: f952e37370eca27953292efe441db0a5ddf0a102
https://github.com/Perl/perl5/commit/f952e37370eca27953292efe441db0a5ddf0a102
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: refactor: make generate_output take Param
The generate_output() method currently takes a hash of args. The fields
of this hash are typically generated on the fly by each caller from the
fields of an ExtUtils::ParseXS::Node::Param object.
This commit makes it so that instead, that Param object is just passed
directly to generate_output(). It simplifies the code, and is a step
along the path to the eventual goal of making generate_output() be a
method of ExtUtils::ParseXS::Node::Param rather than of
ExtUtils::ParseXS.
This commit causes one slight change in behaviour: set magic; and this
is arguably a bug fix.
There are two ways of specifying args that get updated on return (the
XS equivalent of 'sub foo { $_[0]++ }'): either by declaring the parameter
as 'OUT' in the signature:
int
foo(OUT int a)
or by listing it on an OUTPUT line. For the latter, the arg has set
magic called on it after the update by default, but this can be
overridden with the SETMAGIC keyword:
OUTPUT:
arg1 // has a SETMAGIC() call
SETMAGIC: DISABLE
arg2 // doesn't have a SETMAGIC() call
SETMAGIC: ENABLE
arg3 // has a SETMAGIC() call
On the other hand, OUT params have set magic by default *unless*
there's an OUTPUT section with at least one SETMAGIC line, in which case
the last such SETMAGIC line dictates whether the OUT arg gets set magic.
This seems wrong, and this commit changes it so that OUT params *always*
get set magic.
If this turns out to be wrong, the line to look at is
print "\tSvSETMAGIC($arg);\n" if !$param->{in_output} || $do_setmagic;
which says to do magic always for OUT params, and to honour the SETMAGIC
in force at the line for OUTPUT params.
Commit: 45f7c3de30a71f75d0985a948f0dd19fa5dde445
https://github.com/Perl/perl5/commit/45f7c3de30a71f75d0985a948f0dd19fa5dde445
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: fix recent OUTLIST breakage
The last couple of commits broke the unusual (and possibly not very sane)
combination of a parameter declared both as IN_OUTLIST and listed in an
OUTPUT section.
This means that generate_output() is called twice for the same
parameter, once to update the value of the arg's SV on the stack, and
once to push its value onto the stack.
So, distinguish between those two effects by whether the caller of
generate_output() passed an extra arg (for OUTLIST) or not, rather than
by whether the Param object has an in_out field of /OUTLIST$/.
This was manifesting itself as an extra PUSHs and SvSETMAGIC appearing
compared with 5.40.0.
Fix and add tests.
Commit: dfa3e37cdbee2d3907362596c345ac8c7cb349d9
https://github.com/Perl/perl5/commit/dfa3e37cdbee2d3907362596c345ac8c7cb349d9
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
Log Message:
-----------
ParseXS: refactor: simplify RETVAL/TARG code
A recent commit moved, from outside to inside generate_output(), the
perl code which is responsible for generating the C code which handles
the "use TARG if possible when returning RETVAL" optimisation.
This commit now follows that up by simplifying and tidying that code.
Now that it's within generate_output(), the setting of $type, looking up
of the typemap etc, have all already been done at the start of the sub,
so there's no need to duplicate that effort.
Also, set {output_code} as a field value in the faked-up RETVAL
param object, so that inside generate_output() it can just check for
a generic $param->{output_code} field rather than the special-cased
$self->{xsub_RETVAL_typemap_code} value.
Commit: 5c8f70542359469e61041a3577bbd4ad1817a185
https://github.com/Perl/perl5/commit/5c8f70542359469e61041a3577bbd4ad1817a185
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
Log Message:
-----------
ParseXS: refactor: move OUTPUT code handling.
OUTPUT sections can include explicit setting code to override the
normal typemap. For example:
OUTPUT:
arg1 # use standard typemap entry
arg2 my_setiv(ST(1), (IV)arg2); # override
Formerly, the custom code was emitted directly during parsing by
OUTPUT_handler(). This commit moves that emitting code into
generate_output(), so that the emitting with/without override
are done in the same place.
This is a small part of an effort to concentrate all the code-emitting
parts into a small number of well-defined places.
Commit: 9a19c04542b8d833c6d8204b61d6f6df87ae81db
https://github.com/Perl/perl5/commit/9a19c04542b8d833c6d8204b61d6f6df87ae81db
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
Log Message:
-----------
ParseXS: refactor: s/NO_RETURN/NO_OUTPUT/
In earlier refactoring, I renamed an E::PXS object field to
xsub_seen_NO_RETURN, when it should actually have been to
xsub_seen_NO_OUTPUT.
This commit renames the field again.
Should be no functional changes.
Commit: f3e24d547f73ac1312244b59d70a65a5ee23091c
https://github.com/Perl/perl5/commit/f3e24d547f73ac1312244b59d70a65a5ee23091c
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: add test: OUTPUT RETVAL with code
Add a test for the specific:
OUTPUT:
RETVAL template_override_code
Commit: 4d58f44c949d113abb74b64ab89f0d677db7b6a3
https://github.com/Perl/perl5/commit/4d58f44c949d113abb74b64ab89f0d677db7b6a3
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: add tests for RETVAL as a parameter
Normally RETVAL is an implicitly declared local var, but it is legal
to also declare it as an explicit parameter. Add tests for the various
permutations of this:
- void and long XSUB return values;
- no param type; int type in signature; int type in INPUT;
- with and without an autocall.
This isn't well documented as to how it should be interpreted, so these
tests are more about checking current behaviour so that inadvertent
changes are detected, rather than approving the current behaviour.
Some of the current behaviour is less than satisfactory (generally
marked with 'XXX' in the test code comments).
In particular, a commit shortly is going to change how RETVAL is
implemented internally.
It also adds a few tests for NO_OUTPUT.
Commit: 7acbaa629361fe5bb0b204b03b211061c761684e
https://github.com/Perl/perl5/commit/7acbaa629361fe5bb0b204b03b211061c761684e
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Node.pm
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: refactor: add RETVAL Node::Param object
Currently, RETVAL is handled almost entirely using special-case code
all over the place.
For non-void XSUBs, this commit adds a Node::Param object with a name of
RETVAL to the start of the list of Param objects in the Sig object - but
flagged as synthetic and no_init (so it gets declared but not
initialised, and doesn't consume an argument from the stack).
This makes the emitting of the
some_type RETVAL;
declaration be part of general parameter processing, rather than being
special-cased. Soon, this new object will also be used to make emitting
of the code to return RETVAL be more regular; but for now just the
initial declaration is handled in the new fashion.
Note however, that things get complicated when RETVAL is also declared
as a parameter, e.g.
int
foo(RETVAL)
long RETVAL
....
or
int
foo(long RETVAL)
....
or even as a parameter but without a type:
int
foo(RETVAL)
# no INPUT lines
...
or even as an alien variable:
int
foo()
long RETVAL = 99;
Much of the behaviour in such cases is unspecified, and is somewhat
emergent - i.e. it behaves that way due to the way the code happened to
be structured, rather than due to any particular intent. This commit
tries very hard to preserve the current behaviour - hence all the tests
added in the previous commit. It has however changed behaviour in a
couple of places where the old behaviour emitted broken C code. These
changes are indicated by the updates to t/001-basic.t in this commit.
To keep the backwards-compatibility behaviour, the basic approach of
this commit has been to make the RETVAL object be in one of three main
states:
fully-synthetic What is initially pushed at the *start* of the param
list. RETVAL hasn't yet appeared in a signature or INPUT.
Its type is set to the XSUB's return type.
semi-real Same as fully-synthetic, but with a defined
argument number, and with an updated position within
the parameter list. This signifies that a RETVAL has
appeared in the signature, but without a type yet
specified. It acts as a placeholder, causing one passed
argument to be skipped, and to appear in usage() error
messages etc.
real The is_synthetic and no_init flags are turned off and
its type comes from the signature or INPUT line. This
is now just a normal parameter.
Commit: d1d26fab3b52e4bd510b6eb15ad8177d92015a83
https://github.com/Perl/perl5/commit/d1d26fab3b52e4bd510b6eb15ad8177d92015a83
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/t/002-more.t
M dist/ExtUtils-ParseXS/t/XSMore.xs
Log Message:
-----------
ParseXS: refactor: use RETVAL param for output
The previous commit added a synthetic RETVAL Node::Param object and
used it for generating RETVAL declaration code rather than
special-casing it.
This commit extends this to use the object for output processing too,
removing more special-case handling, and removing the need for some
special state fields.
The next few commits will remove more RETVAL special-handling code and
generally simplify and clean up.
Note that this commit actually introduces a bug which will be fixed in
a few commits' time: the new code doesn't correctly handle multiple
CASEs, each with a RETVAL output, such as:
int
foo(...)
CASE: x
...
OUTPUT: RETVAL
CASE: y
...
OUTPUT: RETVAL
The code now sees this as a duplicate OUTPUT parameter. For now, a test
in t/002-more.t and t/XSMore.xs has been disabled until this is fixed.
(Actually, CASE with duplicate INPUT or OUTPUT sections has been broken
for a few weeks - but this is the first commit that actually caused a
test failure. All will be fixed soon...)
Commit: 496574a51ac130393e98f35bf3ccd022a6355473
https://github.com/Perl/perl5/commit/496574a51ac130393e98f35bf3ccd022a6355473
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: use XSUB return type for RETVAL return
Following on from the previous commit, which uses the new RETVAL
Node::Param object for generating output code, ensure that the
return type used for returning the XSUB's value is the XSUB's type
rather than RETVAL's type, where they differ.
For example, with
int
foo(long RETVAL)
RETVAL is is declared as long, but an int is returned to the caller.
Add tests to confirm.
Commit: ffc833cfeb83e93399917b83fb7c19db69a5d963
https://github.com/Perl/perl5/commit/ffc833cfeb83e93399917b83fb7c19db69a5d963
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: refactor: better handle NO_OUTPUT
In an XSUB like
NO_OUTPUT int
foo(...)
An 'int RETVAL;' declaration is emitted, but no code is emitted to
return the value of RETVAL to the caller.
Before this commit, the way this was handled was that, after all INPUT
and CODE/PPCODE/autocall code had been emitted, ParseXS.pm did the
following:
($implicit_OUTPUT_RETVAL, $self->{xsub_return_type}) =
(0, 'void') if $self->{xsub_seen_NO_OUTPUT};
That is to say, halfway through parsing the XSUB, it suddenly pretends
that the XSUB was declared void all along, and that there wasn't an
autocall (which would give an implicit OUTPUT: RETVAL declaration).
While this works, it is a confusing action-at-a-distance: at any
particular point in the code, you don't easily know whether
if ($self->{xsub_return_type} eq 'void) { ... }
is dealing with a real void or fake void.
This commit removes that halfway-along state modification, and instead
handles NO_OUTPUT directly where needed:
For $implicit_OUTPUT_RETVAL, it now just never sets it if NO_OUTPUT is
present, rather than setting it and then later resetting it.
For $self->{xsub_return_type}, this now always represents the actual
declared return type of the XSUB, and whenever necessary,
$self->{xsub_seen_NO_OUTPUT} is checked along side
$self->{xsub_return_type} in various places.
In addition, the $var_num variable in OUTPUT_handler() handler has been
removed, as it was a vestigial value, now only used to determine whether
to call
$self->blurt("Error: OUTPUT $outarg not a parameter")
which can be done directly as boolean expression.
Several new NO_OUTPUT tests have been added.
Commit: bbb99046c5bcdf162386cb9f80584af79dab3a8a
https://github.com/Perl/perl5/commit/bbb99046c5bcdf162386cb9f80584af79dab3a8a
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
Log Message:
-----------
ParseXS: refactor: reindent block
3 commits ago I added a new scope. This commit re-indents that new block.
Whitespace-only (apart from deleting a stray empty comment line too).
Commit: a9b73fc1e600b07bf927f0def4ea3555550d0164
https://github.com/Perl/perl5/commit/a9b73fc1e600b07bf927f0def4ea3555550d0164
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Node.pm
Log Message:
-----------
ParseXS: refactor: simplify ST() method.
The $self->ST($num) method typically emits "ST(" . ($num-1) . ")"
but it has special-case handling for $num being undef, or (in the case
of output), being 0.
The latter was a special-case for outputting RETVAL return code. Recent
refactoring means that the ST() method is no longer called with
$num == 0, so the method's code can be simplified (and the $is_output
parameter eliminated).
Should be no change in functionality.
Commit: b826448d5bbbef9e210302bd88571c59184af72f
https://github.com/Perl/perl5/commit/b826448d5bbbef9e210302bd88571c59184af72f
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: improve an error message
For the specific case of RETVAL appearing in an OUTPUT section when
NO_OUTPUT is in force, output this new more specific error message:
can't use RETVAL in OUTPUT when NO_OUTPUT declared
rather than the more generic (and in this case, slightly confusing)
message:
OUTPUT RETVAL not a parameter
Commit: 88bb21ac00b04a1626b1616c321cd587ff502de1
https://github.com/Perl/perl5/commit/88bb21ac00b04a1626b1616c321cd587ff502de1
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Node.pm
M dist/ExtUtils-ParseXS/t/001-basic.t
M dist/ExtUtils-ParseXS/t/002-more.t
M dist/ExtUtils-ParseXS/t/XSMore.xs
Log Message:
-----------
ParseXS: fix CASE:
My recent merge commit v5.41.4-108-g9621dfa822 partially broke the CASE:
functionality; although I didn't spot it at first since the ParseXS test
suite only has a single CASE: test, and none of the XS code bundled with
perl core makes use of it.
The issue was that an XSUB with multiple CASEs has a single signature,
but potentially multiple sets of INPUT and OUTPUT sections. For example:
int
foo(abc, def)
CASE: X
int abc;
int def;
CODE:
RETVAL = abc + def;
OUTPUT:
RETVAL
CASE: Y
long abc;
long def;
CODE:
RETVAL = abc - def;
OUTPUT:
RETVAL
Once my merge commit made the data from INPUT and OUTPUT sections be
used to update each Node::Param object (which has been derived from the
signature entry) then this starts to fall apart. It only started
failing the single CASE test in t/002-more.t once *this* branch
made RETVAL an object (rather than it being special-cased).
The fix in this commit is a bit of hack. Once all of the XSUB's
signature has been parsed into a list of Node::Param objects, a copy of
this list is taken and kept as the original. Then each time the body is
parsed (typically once, but multiple times for bodies with CASE
keywords) that original list is copied and becomes the current parameter
list. The objects in that list are updated by INPUT and OUTPUT sections
as usual. Then for the start of the next CASE, the next copying blows
away all those updates.
If and when ParseXS takes on more of an AST-like appearance (with,
hypothetically, Node::INPUT, Node::OUTPUT etc objects) then this
approach should be revisited.
Note that this commit restores the CASE test in t/002-more.t which I
temporarily disabled a few commits ago.
Commit: 8eaf72eca85eb112b7ed4d90dac3655eee8b5d9f
https://github.com/Perl/perl5/commit/8eaf72eca85eb112b7ed4d90dac3655eee8b5d9f
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Node.pm
Log Message:
-----------
ParseXS: refactor: reindent Node::Sig::parse()
My previous merge commit moved (and renamed) this function from
ParseXS.pm to ParseXS/Node.pm. The former file has an indent of 2, while
the latter has an indent of 4. When I moved the code across, I forgot to
reindent it. So this commit does that now.
Whitespace-only change, apart from rewrapping some code comment blocks.
Commit: a67a29a84f6e8e0fcaffb35d0f47332653440851
https://github.com/Perl/perl5/commit/a67a29a84f6e8e0fcaffb35d0f47332653440851
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/t/001-basic.t
Log Message:
-----------
ParseXS: t/001-basic.t: fixup regex
A couple of newly-added regexes had (in part) /....\b*/, which gives the
warning:
\b* matches null string many times in regex
so this commit removes the extraneous '*'
Commit: 6644e0029afa5855c52207342e037b9ba08bc93d
https://github.com/Perl/perl5/commit/6644e0029afa5855c52207342e037b9ba08bc93d
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
Log Message:
-----------
ParseXS: fix under older perls
One line had: 'exists $param->{type}' as part of a condition. This
worked fine when $param was a normal hash ref (created under 'use
fields'), but the combination of:
- a recent commit which copies those hashes to $sig->{old_params}
then resets some of the fields (to allow for multiple CASEs);
- older perls (in particular 5.8.9) which used pseudo-hashes to
implement fields,
was causing the field to spring into existence, even though it wasn't defined.
So change the test from 'exists' to 'defined'.
Commit: a1fad1eeb8b565d675b85b5eb3832bbfb97f78be
https://github.com/Perl/perl5/commit/a1fad1eeb8b565d675b85b5eb3832bbfb97f78be
Author: David Mitchell <da...@iabyn.com>
Date: 2024-11-18 (Mon, 18 Nov 2024)
Changed paths:
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS.pm
M dist/ExtUtils-ParseXS/lib/ExtUtils/ParseXS/Node.pm
Log Message:
-----------
ParseXS: refactor: use new() copying to simplify code
A recent commit needed, at one point, to make a copy of a Node::Param
object, and did so in a clunky way. But this can be simplified by
instead using the new() method's built-in ability to take an
initialisation hash ref as an arg. So this commit does that.
However, it turns out that this new() method *shared* rather than copied
that hash ref when running in an environment where 'use fields' isn't
available - namely during the building of perl itself. So fix that too.
Compare: https://github.com/Perl/perl5/compare/527a03f59d2c...a1fad1eeb8b5
To unsubscribe from these emails, change your notification settings at
https://github.com/Perl/perl5/settings/notifications
