Changes

Fix INRIA CI job other-configs

usage warnings for private constructors: document our intentions in the testsuite

document why sprintf still exists despite asprintf

rewrite doc comment

update comment according to @octatron's suggestion

fix local variables in Makefile

Fix C++ name-mangling issue with `caml_state` on Cygwin (#14220)

Fix Cygwin problems linking with caml_state from C++

---------

Co-authored-by: Antonin Décimo <antonin@tarides.com>
Co-authored-by: David Allsopp <david.allsopp@metastack.com>

fixes suggested by reviewer

Update stdlib/format.mli

Co-authored-by: Antonin Décimo <antonin.decimo@gmail.com>

Disable Tsan on one more deliberately-racy test. (#14232)

Manual attributes page fixes (#14233)

* Change quotes to display attributes names as inline code blocks

* Fix typos

* Reflow text

Fix ocamltest criterion for using `-custom` to link test file. (#14230)

Ocamltest uses the wrong criterion when deciding whether to link the
test program with `-custom`.

Instead of looking at the libraries being linked and checking them
for dynamically-linked C code, simply check their `lib_custom` flag.

manual(etex): trim leading spaces (#14228)

Fix test-in-prefix (#14216)

Add `flexlink.exe` and `ocamllex.exe` to the list of files that do not
link with the Config module. Normally, these are symlinks to `*.byte.exe`
or `*.opt.exe` but (depending on Windows shenanigans) `configure` may
decide to use `cp` instead of `ln -s`.

Integral type hygiene in compaction (and shared_heap.c generally).

Avoid an undefined variable warning from Make

Co-authored-by: David Allsopp <david.allsopp@metastack.com>

parsetree: map on all locations in toplevel directives

typing: allow private constructors without usage warnings

We would like to allow the use of private variant constructors to
build 'phantom' types used for GADT indices:

    type zero = private Zero
    type 'a succ = private Succ

    type ('a, _) sized_list =
    | Nil : ('a, zero) sized_list
    | Cons : 'a * ('a, 'n) sized_list -> ('a, 'n succ) sized_list

Before the present commit, this example would fail with "unused
constructors" warning on Zero and Succ.

(We still warn on type constructors that are defined as public, and
then exported as private in a signature. But on types that are defined
as private in the implementation, warning that they are not used is
not that useful anyway, because it is impossible to construct them
except through the FFI.)

Note: we could reasonably warn on constructors that are defined as
private in the implementation, and then used in pattern-matching (but
never constructed, as constructing them is a type error). But this is
arguably warning about something different, and there is a risk of
negatively affecting existing projects by raising unused-constructor
warnings more often.

Reviewed-by: Florian Angeletti <florian.angeletti@inria.fr>
Reviewed-by: Nicolás Ojeda Bär <n.oje.bar@gmail.com>

Changes

Add native debugger manual section

Fixup name mangling references.

Don't use variadic macros for annotations

Varadic macros expect at least one parameter for the variadic part (at
least until C++20). They were used to allow for possible future
extension by adding more parameters.

    /Users/antonin/Tarides/ocaml/trunk/runtime/caml/memory.h:89:32: error: ISO C++11 requires at least one argument for the "..." in a variadic macro [-Werror]
       89 | CAMLmalloc(caml_stat_free, 1, 1) CAMLreturns_nonnull()
          |                                ^

Changes

GHA: update from checkout@v4 to checkout@v5

GHA: update Debian i386 image

explain the syntax change for variance (by #13820) (#14126)

GHA: factor Android NDK et API level numbers

Fix the CI: Skip testsuite check on AppVeyor

Report mingw-runtime version on x86_64 as well

AppVeyor reports the version of the mingw-i686-runtime package, but that
wasn't copied over when the other ports were added on AppVeyor.

Fix make runtop on Windows

GHA: fold config.log output in failed cross-builds

Turn off the echo'ing of command with `set +x` when displaying
`config.log`, otherwise it may scramble the output.

No `pthread_cancel` in Android's Bionic libc

Replacing `pthread_cancel` with assert using the preprocessor doesn't
work if `assert` isn't defined either! It was `<assert.h>` that was
missing, prefer detecting at configure-time whether `pthread_cancel`
is supported, and guard its use. Another option would be to guard its
use using the `__BIONIC__` macro.

- https://android.googlesource.com/platform/bionic/+/master/tests/headers/posix/pthread_h.c#90
- https://stackoverflow.com/questions/26823926/pthread-cancel-not-found-compiling-with-android-toolchain

Misc. Autoconf simplifications

stdlib: add filter_mapi function (#14227)

Add caml_check_error_exn

Preparation: caml_thread_new_descriptor_exn

Preparation: install_backup_thread_exn

We move install_backup_thread_exn earlier to avoid having to undo
the domain creation.

[clean-up] Remove unused field

[minor] Factor domain_thread_func

Handle errors during domain spawn

Call non-raising variants of lock/broadcast

Changes

reviewer suggestions

parsing: refactor ast {iter,map} location functions (#14237)

document domains joining their systhreads before being joined (#14248)

Allow strings to be shared by Lambda.make_key

Strings have been immutable for a while now

Use Const_immstring for all constant strings created by the compiler

Introduce Lambda.lambda_of_const

This helper function ensures that constant strings are always
represented using Const_immstring and not Const_base (Const_string _)

Introduce Lambda.constant type

This is a copy of Asttypes.constant minus the string case.
It ensures that there is a single way to represent constant strings,
and allows to remove a few pieces of duplicated code.

Remove Lambda.Const_base indirection

Changes for PR14260

Fix TSan shadow stack popping on some C stubs (#14213)

Until now, when an exception was raised from C and TSan was enabled, the
runtime would unwind the (real) stack and pop on TSan's shadow stack at
every frame, until the stack pointer was at a higher address than the
c_stack_link. This is incorrect: the correct condition is to stop when
the stack pointer is no longer between the c_stack_link and the stack
pointer's value before unwinding. This is because unwinding beyond the
c_stack_link means going back to an OCaml stack, which is on the heap,
and may be stored at an arbitrary address, higher or lower than the C
stack.

This change also makes the code clearer by using a for loop and
improving comments.

Co-authored-by: Gabriel Scherer <gabriel.scherer@gmail.com>

manual: always refer to headers as <caml/example.h>

Unicode changes in Format should be treated as breaking changes (#14272)

See https://github.com/dbuenzli/cmdliner/issues/234 where this potentially causes a testcase that was previously passing on 5.3 to now fail on 5.4.0~beta2.

stdlib: implement chop_suffix_opt in term of check_suffix

stdlib: no allocation for Filename.is_implicit

manual: add explanation about 'simple' GC rules (#14273)

* manual: add explanation about 'simple' GC rules

* Review

Remove unused variable from fp_backtrace (#14277)

dsource: preserve (mod) whenever this form is allowed

Fix a typo in the Repr.compare documentation. (#14281)

Changes entry

Typos, typos (#14283)

Use multicoretests.0.10 compatible with restored Gc.stat stack_size ocaml-multicore/multicoretests#567

Bugfix for GC root mishandling in memprof.c

When allocating the OCaml block for a Gc.Memprof info block, memprof.c uses
e->user_data as the root for the callstack. While e->user_data is indeed a root,
this is incorrect: e->user_data is only scanned during minor GC if it is the
right side of the 'young' pointer, and this might not be the case.

The fix is to use a normal CAMLlocal1 root for this value.

The testcase is tricky, but reliably crashes before and passes after this fix.

Changes

Use multicoretests.0.11 compatible with adjusted Domain.spawn error msg ocaml-multicore/multicoretests#571

update info in preparation of the 5.4.0 release

sync Changes with 5.4.0

update calendar

Add caml_find_index_of_running_domain

Author:    Gabriel Scherer <gabriel.scherer@gmail.com>
Co-authored-by: Jack Nørskov Jørgensen <j@nojo.dk>

Add caml_c_thread_register_in_domain

Add tests for caml_c_thread_register_in_domain

Add caml_c_thread_register_in_domain to manual

Add Changes entry

Display more debugging information

Avoid TLS in st_stubs.c

Use Digest.BLAKE128 for the compiler's consistency checks

Bootstrap

Formatting

Update locations

Changes

Improve doc of Runtime_events.Timestamp

Mention changes in Changes

improve more

update Changes with reviewer info

.gitmodules: add the branch config entry

It makes it clearer how to select a different branch than master when
experimenting.
The file doesn't need a copyright header.

Mark FLEXLINKFLAGS as precious and pass them to flexlink invocation

Flexlink reads the FLEXLINKFLAGS env var for supplementary flags. They
could be used when building the compiler. If we don't add them
explicitly to `$flexlinkflags` they're:

1. not shown in the flexlink invocations;
2. not retained after the build, when the installed ocaml calls
   flexlink as a linker.

Prefer passing the FLEXLINKFLAGS explicitly in the invocation, at the
risk of duplicating the flags if FLEXLINKFLAGS is set in the
environment rather than given to configure as an argument.

This change also marks the variable as precious so that any change to
it invalidates the build.

ocaml-variants.opam: require FlexDLL >= 0.44

FlexDLL before 0.44 would try to rediscover the C compiler and other
tools it should use. This would introduce a mismatch if users chose an
C compiler when building OCaml, and FlexDLL would chose another
compiler.

As of FlexDLL 0.44, its Makefile exposes all of the required
settings. OCaml build scripts now know how to forward the selected
tools to FlexDLL build scripts. Update the lower bound to set that in
stone.

Define RC to choose an alternative resource compiler

This allow overriding the default resource processor, used during
FlexDLL's bootstrap. The variable name is defined by libtool.

Pass the variable to the flexlink Makefile.

Pass $CC in the flexdll build env

FlexDLL's Makefile has its own mechanism to discover the C
compiler. Override it with the selected C compiler.

Update changes

Use clang-cl for flexdll support objects

Don't persist FLEXLINKFLAGS in flexlink_flags

Define MANIFEST_TOOL to choose an alternative manifest generator

This allows using `llvm-mt` instead of `mt.exe`. The MANIFEST_TOOL
variable is defined by libtool.

tweak documentation of Runtime_events.Timestamp

Amend the documentation of the constraint

Have a less committal description to leave the door open to possible
future changes to this constraint

Make the type checker more deterministic (#14297)

* Typetexp: require non-randomized hash tables to produce deterministic types

* Use maps instead of hash tables to ensure deterministic iteration order

* Hygiene

* Add comments

Modify frame pointer tests to normalise symbols

Each platform has a different name mangling scheme, however for these
tests we can just replace the '$' with '.' normalising the names
across platforms. These tests don't require escaping of special
characters so it is safe to do this.

Translobj: ensure consistent order for placing shared constants (#14299)

Add missing TSan accounting in caml_c_call_stack_args for x86_64

The absence of these calls to TSan resulted in its shadow stack being
offset and deriving over time, causing wrong backtraces and crashes (due
to the complete absence of bound checks in TSan).

Fix caml_c_call_stack_args for ARM64, POWER, RISC-V and s390x

Update TSan tests after #12410

Update changelog

fix typo in ci action (#14309)

I didn't add the reviewers for #13785, apologies.

Changes: add missing 'and' and ','

Always use the `"I"` format specifier for `size_t` with mingw (#14308)

* Always use the "I" format specifier for size_t with mingw

    runtime/domain.c:520:7: error: unknown conversion type character 'z'
    in format [-Werror=format=]
      520 |       "young_start: %p,"
          |       ^~~~~~~~~~~~~~~~~~
    In file included from runtime/domain.c:22:
    runtime/caml/config.h:71:35: note: format string is defined here
       71 | #define ARCH_SIZET_PRINTF_FORMAT "z"
          |                                   ^

In some configurations GCC will warn about "%z" not being supported.
Partially revert a change introduced in
295dd45852688c7c8b5df5e974c460588cdd9696.

> Some types are different sizes in 32-bit and 64-bit code. For
> example, size_t is 32 bits long in code compiled for x86, and 64
> bits in code compiled for x64. To create platform-agnostic
> formatting code for variable-width types, you can use a
> variable-width argument size modifier. Instead, use a 64-bit
> argument size modifier and explicitly promote the variable-width
> argument type to 64 bits. The Microsoft-specific I (uppercase i)
> argument size modifier handles variable-width integer arguments, but
> we recommend the type-specific j, t, and z modifiers for
> portability.

> Microsoft-specific:
> The I (uppercase i) […] argument size modifier prefix [is] Microsoft
> extension and is not ISO C-compatible.

https://learn.microsoft.com/en-us/cpp/c-runtime-library/format-specification-syntax-printf-and-wprintf-functions?view=msvc-170#size

Presumably the GCC warning is outdated or erroneous, but there seem to
be no way around it.

* Update changes

Detect DWARF version for the Clang assembler (or: fix `test-in-prefix` for the other-configs tests) (#14314)

This is then used in testsuite/tools/toolchain.ml to correctly identify
whether the clang assembler embeds the full path to the source file or
not.