This reverts commit 5f02d0d9e1.
Which isn't correct, we have very strange behavior of "force on pie/pic
or let the toolchain do whatever it wants, but you can't turn it off."
Calling `isinstance(self, X)` is an anti-pattern, we should just be using
inheritance for this, letting the `StaticLibrary` override the method, and
having the base class always return `False`.
Since they are actually dependencies out the output not the Generator
itself.
This fixes dependency issues in the ninja backend, allowing Meson to
rebuild more accurately. It also does sometimes in the vs backend, but
there are problems in the vs backend I'm not sure how to solve. The
vsbackend is, itself, so fragile looking I don't want to get too
involved with it.
This doesn't actually fix the problem, but it provides parity with what
is currently happening. I don't have access to a Windows machine to
further debug, however, so not breaking anything is the best I can do
ATM.
Leak sanitizer can be enabled without the whole AddressSanitizer, this
can be done by passing -fsanitize=leak as documented at [1].
Meson doesn't support this, so add support for it.
[1] https://clang.llvm.org/docs/LeakSanitizer.html
We might be using all fallbacks, or be super weird and not use
bash-completion, or simply have a different PKG_CONFIG_LIBDIR set. And
devenv already checks whether the dependency is found, but defaults to
required anyway, which is wrong.
Currently, the code puts a placeholder in for the first output, then
replaces all of the outputs when it generates final value. Instead,
let's only replace the placeholder value.
Fix "Tried to grab file outside current (sub)project" error when subproject exists within
a source tree but it is used through a symlink. Using subprojects as symlinks is very useful
feature when migrating an existing codebase to meson that all sources do not need to be
immediately moved to subprojects folder.
Logically, i18n.merge_file cannot ever take a MULTI_OUTPUT_KW, but it
does take a CT_OUTPUT_KW-like interface.
Actually trying to pass multiple merge_file outputs causes the
msgfmthelper script to be entirely malformed in the arguments it
accepts, and treat the broken one like a --flag, then exit with argparse
errors.
Even if we somehow assumed that somehow it was designed to actually
allow this, msgfmt doesn't support conceptually passing multiple outputs
so that would be a msgfmt error instead of an error inside the guts of
`meson --internal msgfmthelper`.
Same logic applies again for the itstool command and the itstool
internal helper.
Catch this error at configuration time by using the single-output kwarg
form.
Likewise, it's totally nonsense to accept multiple install_dir or
install_tags, and ever since commit 11f9638035
the CustomTarget itself won't even check this.
CT_OUTPUT_KW is the same OUTPUT_KW we use in lots of places. The most
distinctive thing about it is not that it's part of custom_target
(basically any other function that uses such a kwarg follows the same
rules due to using CustomTarget under the hood), but the fact that it
takes multiple outputs.
We validate a few things here, such as the non-presence of '@INPUT' in
an output name. These got moved out of the CustomTarget constructor in
commit 11f9638035 and into KwargInfo, but
only for kwargs that took multiple values. This caused configure_file()
and unstable_rust.bindgen() to stop checking for this.
Add a shared single-output KW and use it in both places. This now
dispatches to _output_validator.
configure_file now validates subdirectories in output names the same way
we do elsewhere, directly in the typed_kwargs and by specifying the
erroring kwarg.
Or any other reserved names. We check in add_target that the primary
name of any build target isn't on the forbidden list, but custom_target
allows names that are distinct from the output filenames, so we need to
check those too.
We would eventually still error out all the way at the end, with:
```
ERROR: Multiple producers for Ninja target "all". Please rename your targets.
```
But, if we can check that early and provide the underlying reason
(reserved name) alongside actually useful debugging info (a line
number), then why not?
Refactor the check into a small helper function in the process.
We don't want to allow targets that conflict with:
- our aliased meson-* targets for phony commands
- any meson-*/ directories we create for internal purposes
We do want to allow targets such as:
- our own meson-*.X manpages
There are a couple routes we could take.
Using a better restriction, such as `meson-internal__*`, is trivially
done for our aliased targets, but changing directory names is...
awkward. We probably cannot do this, and doing the former but not the
latter is not very useful.
We could also carefully allow patterns we know we won't use, such as
file extensions, but which the manpages need, which works for our
directories and for many aliased targets, but run_target() is
user-specified and can be anything.
Use a hybrid approach to cover both use cases. We will now allow target
names that fulfill *all* the following criteria:
- it begins with "meson-"
- it doesn't continue with "internal__"
- it has a file extension
Every phony target has a special indirection rule created because ninja
is bad at deleting generated outputs and tries to delete phony outputs
too.
Instead of invoking this as a separate helper post-creation function to
create the alias, wrap NinjaBuildElement and create it behind the
scenes. This simplifies target naming and means one less line at every
single use site.
The `install_headers` function now has an optional argument
`preserve_path` that allows installing multi-directory
headerfile structures that live alongside sourcecode with a
single command.
For example, the headerfile structure
headers = [
'one.h',
'two.h',
'alpha/one.h',
'alpha/two.h',
'alpha/three.h'
'beta/one.h'
]
can now be passed to `install_headers(headers, subdir: 'mylib', preserve_path: true)`
and the resulting directory tree will look like
{prefix}
└── include
└── mylib
├── alpha
│ ├── one.h
│ ├── two.h
│ └── three.h
├── beta
│ └── one.h
├── one.h
└── two.h
Fixes#3371
We have two checks for the type accepted here. One is the basic
typed_kwargs type-checking, which declares it accepts:
str | include_directories
The other is the custom validator which further narrows it to strings
with certain option-like properties (needs to be an = assignment).
The former is obviously wrong, which doesn't really matter all that much
but still isn't very nice...
Introduced in commit f34013fb08.
override_options makes no sense for custom_target as we don't use it for
anything. Also, this was added in commit c3c30d4b06
despite not being allowed in permittedKwargsc3c30d4b0.
For inexplicable reasons, we had a known_kwargs for custom_target that
looped over kwargs and issued a warning, not an error, for unknown
kwargs. It was impossible to ever hit that check to begin with, though,
ever since commit e08d735105 which added
permittedKwargs and obsoleted those manual checks with real errors.
So at one point override_options was specially permitted to be used
without emitting a warning, and then for about half a decade it was an
error, and then based on some dead code it was allowed again for a bit.
But through all this it doesn't do anything and isn't documented.
For maintainer targets, we need some more tools that gettext-tiny
doesn't implement. It's a shame to cause NLS to be completely disabled
in such environments, so instead just issue a warning and continue.
Before 0.62.0 these were never checked for, and would simply fail at
runtime, probably. In theory, the user might install the tools in
between configuring and building, and then the maintainer targets would
begin to work. Return to that behavior -- we still create the targets,
which will *probably* fail, but might not -- and for existing
integrations, failing at `ninja foo-update-po` with "error, program
msgmerge not found" is a bit more discoverable than ninja saying "what
do you mean, there's no such target".
We still have the 0.62.0 preferred behavior of trying to find the
programs, succeeding in all cases other than gettext-tiny, and
guaranteeing that their paths are set up in a machine-file-respecting
manner.
In the former case, the presence of tools is optional, but triggers a
warning and then no-ops the target. In the latter case, the presence of
the tools is mandatory. But if it was already looked up and discovered
to be missing, we did not actually check that it is found before trying
to use it.
In the case that it isn't found, check again, so that we explicitly
error out with a relevant error message due to setting the required
flag.
Fixes#10320
mingw GCC using ld.bfd emits diagnostics that include
"-plugin-opt=-pass-through=-lmoldname" and this triggers a match for
mold when it should not.
Instead, always check the very beginning of the output for the linker
id. This is pretty consistent:
- it is always on stdout
- GCC may put additional things on stderr we don't care about
- swift is bizarre and on some OSes redirects the linker stdout to
swiftc's stderr, but it will still be the only stderr; we didn't even
check stderr at all until commit 712cbe0568
For gold/bfd linkers, the linker id is always the *second* word, after
the legally mandated "GNU" we already check for.
Dylan Baker
Paweł Marczewski
Marco Trevisan (Treviño)
Remi Thebault
Eli Schwartz
Xavier Claessens
Vili Väinölä
Karl Linden
Florian "sp1rit"