Dear OCaml users,

We have the pleasu= re of celebrating the birthday of Inge Lehmann by announcing the
release= of OCaml version 5.2.0.

Some of the highlights in OCaml 5.2.0 are:<= /div>

- Re-introduced GC compaction
GC compaction can now= be manually triggered by calling `Gc.compact ()` manually.
This = is expected to be particularly useful for programs that wish to release mem= ory
to the operating system after a temporary memory-intensive ph= ase.

- Restored native backend for POWER 64 bits
Wi= th this restored backend, all 64 bits architecture supported in OCaml 4 are=
supported bin OCaml 5

- Thread sanitizer support
= ; Thread sanitizer is a dynamic data race detector which instrument memory<= br> accesses to detect and explain data races at execution time. Sinc= e the
instrumentation is costly (with a 2x to 7x slowdown), it mu= st be enabled with
the `ocaml-option-tsan` configuration flag. (T= he reference manual contains
more information on how to use TSAN.= )

- New Dynarray module
This new standard library module p= rovides a standard implementation for resizeable array,
which is = guaranteed to be memory safe even in presence of data races.

- New -= H flag for hidden include directories
This new flag makes it poss= ible for build tools to split cleanly dependencies
between direct= (the dependencies explicitly added by the project) and indirect
= dependencies (the dependencies introduced by the direct dependencies) witho= ut
the quirks of previous implementations.

- Project-wide = occurence metadata support for developer tools
When compiling a m= odule with the `-bin-annot` and `-bin-annot-occurrences`
flags, t= he compiler stores in the `.cmt` file an index of all occurences of
&nbs= p; values, types, modules, ...

- Raw identifiers
To improv= e OCaml upward-compatibility, there is a new syntax for lowercase
= identifiers, `let \#if =3D 0`, which works even if the identifier is a key= word
in some OCaml versions. This change has been adopted in OCam= l 5.2 in
preparation of the introduction of the `effect` keyword = in OCaml 5.3

- Local open in type expressions
Local open a= re now allowed in type expression: `val (+): Int64.(t -> t -> t)`.
And a lot of incremental changes:

- Around 20 new functi= ons in the standard library besides the new Dynarray module
(in t= he `Array`, `Float`, `Format`, `Fun`, `In_channel`, `Out_channel`, and `Ran= dom` modules )

- Many fixes and improvements in the runtime
- Man= y bug fixes

OCaml 5.2.0 is still a somewhat experimental release com= pared to the OCaml
4.14 branch. In particular

   - The = Windows MSVC port is still unavailable.
   - Ephemeron perform= ances need to be investigated.
   - `statmemprof` is being tes= ted in the developer branch of OCaml.
   - There are a number = of known runtime concurrency or GC performance bugs
   &n= bsp; (that trigger under rare circumstances).

Since the Windows MSVC= port and statmemprof are still missing, the maintenance
support for OCa= ml 4.14 will be extended until at least the end of the year.

Please = report any unexpected behaviours on the OCaml issue tracker (https://github= .com/ocaml/ocaml/issues)
and post any questions or comments you might ha= ve on our discussion forums (https://discuss.ocaml.org).

The full li= st of changes can be found in the changelog below.

Happy hacking,=E2=80=94 Florian Angeletti for the OCaml team.

Installation Instru= ctions
------------------------------

The base compiler can be in= stalled as an opam switch with the following commands:

  &= nbsp; opam update
    opam switch create 5.2.0

The= source code for the release candidate is also directly available on:
* GitHub: https://github.com/ocaml/ocaml/archive/5.2.0.tar.gz
* OCaml = archives at Inria: https://caml.inria.fr/pub/distrib/ocaml-5.2/ocaml-5.2.0.= tar.gz

Fine-Tuned Compiler Configuration
------------------------= ----------------------

If you want to tweak the configuration of the= compiler, you can switch to the option variant with:

  &n= bsp; opam update
    opam switch create <switch_name&g= t; ocaml-variants.5.1.0+options <option_list>

where `<optio= n_list>` is a qpace separated list of `ocaml-option-*` packages. For ins= tance, for a `flambda` and `no-flat-float-array` switch:

  = ; opam switch create 5.2.0+flambda+nffa ocaml-variants.5.2.0+options = ocaml-option-flambda ocaml-option-no-flat-float-array

Changes in OCa= ml 5.2.0 (13 May 2024)
------------------------------------

(Chan= ges that can break existing programs are marked with a "*")

### Rest= ored backends:

- #12276, #12601: native-code compilation for POWER (= 64 bits, little-endian)
(Xavier Leroy, review by KC Sivaramakrish= nan, Anil Madhavapeddy,
   and Stephen Dolan)

- #12667:= extend the latter to POWER 64 bits, big-endian, ELFv2 ABI
(A. Wi= lcox, review by Xavier Leroy)

### Runtime system:

- #12193: R= e-introduce GC compaction for shared pools
Adds a parallel compac= tor for the shared pools (which contain major heap
blocks sized l= ess than 128 words). Explicit only for now, on calls to
`Gc.compa= ct`.
(Sadiq Jaffer, Nick Barnes, review by Anil Madhavapeddy, Dam= ien Doligez,
   David Allsopp, Miod Vallat, Artem Pianykh, Ste= phen Dolan, Mark Shinwell
   and KC Sivaramakrishnan)

-= #12114: Add ThreadSanitizer support
(Fabrice Buoro and Olivier N= icole, based on an initial work by Anmol Sahoo,
   review by D= amien Doligez, S=C3=A9bastien Hinderer, Jacques-Henri Jourdan, Luc
= ; Maranget, Guillaume Munch-Maccagnoni, Gabriel Scherer)

-= #12850: Update Gc.quick_stat data at the end of major cycles and compactio= n
This PR adds an additional caml_collect_gc_stats_sample_stw to = the major heap
cycling stw. This means that Gc.quick_stat now act= ually reflects the state of
the heap after a major cycle or compa= ction.
(Sadiq Jaffer, review by Miod Vallat and Gabriel Scherer)<= br>
- #12859: Ensure Gc.compact does a full major before the compactor r= uns
(Sadiq Jaffer, review by Leo White, Mark Shinwell, Gabriel Sc= herer,
   Josh Berdine, David Allsopp and KC Sivaramakrishnan)=

- #10111: Increase the detail of location information for debugging= events to
allow the end line number and character offset to be r= eported.
(David Allsopp, review by Nick Barnes, Enguerrand Decorn= e and Stephen Dolan)

- #10403, #12202: introduce `caml_ext_table_add= _noexc` that does not
raise `Out_of_memory` exceptions and use it= inside the blocking sections
of `caml_read_directory`. Als= o, check for overflows in ext table sizes.
(Xavier Leroy, report = by Arseniy Alekseyev, review by Gabriel Scherer)

- #11332, #12702: m= ake sure `Bool_val(v)` has type `bool` in C++
(Xavier Leroy, repo= rt by ygrek, review by Gabriel Scherer)

- #12772, #12787: Avoid usin= g _Bool in public headers for the sake of C++
compatibility
&n= bsp; (Guillaume Munch-Maccagnoni, report by KC Sivaramakrishnan, review
= by Xavier Leroy and KC Sivaramakrishnan)

- #12223: Constify c= onstructors and flags tables in C code. Now these
tables will go = in the readonly segment, where they belong.
(Antonin D=C3=A9cimo,= review by Gabriel Scherer and Xavier Leroy)

- #12234: make instrume= nted time calculation more thread-safe on macOS.
(Anil Madhavaped= dy, review by Daniel B=C3=BCnzli and Xavier Leroy)

- #12235, #12468:= introduce and use the `CAMLnoret` macro as
a lighter alternative= to `CAMLnoreturn_start` / `CAMLnoreturn_end`.
Implement it so as= to conform with C11, C23, C++11, C++17.
(Xavier Leroy and Dhruv = Maroo, with help from Antonin D=C3=A9cimo, review by
   Gabrie= l Scherer and David Allsopp)

- #12275: caml/stack.h: more abstract m= acros to describe OCaml stacks and
how to traverse them, supporti= ng more stack layouts.
(Xavier Leroy, review by KC Sivaramakrishn= an and Fabrice Buoro)

- #12268: deliver `Out_of_memory` exception if= domain creation fails
due to memory resource exhaustion. I= t was previous always a `Failure`.
(Anil Madhavapeddy, review by = David Allsopp)

- #12300, #12314: Discard out_channel buffered data o= n permanent I/O error
(Xavier Leroy, report by T=C3=B6r=C3=B6k Ed= win, review by Anil Madhavapeddy
and Nicol=C3=A1s Ojeda B=C3=A4r)=

- #11386: Simplifications and fixes to multicore systhreads impleme= ntation.
(Guillaume Munch-Maccagnoni, review by Anil Madhavapeddy= and KC
   Sivaramakrishnan)

- #12875, #12879, #12882: = Execute preemptive systhread switching as a
delayed pending actio= n. This ensures that one can reason within the
FFI that no mutati= on happens on the same domain when allocating on
the OCaml heap f= rom C, consistently with OCaml 4. This also fixes
further bugs wi= th the multicore systhreads implementation.
(Guillaume Munch-Macc= agnoni, bug reports and suggestion by Mark
   Shinwell, review= by Nick Barnes and Stephen Dolan)

- #12408: `Domain.spawn` no longe= r leaks its functional argument for
the whole duration of the chi= ldren domain lifetime.
(Guillaume Munch-Maccagnoni, review by Gab= riel Scherer)

- #12409: Fix unsafety and deadlocks should an asynchr= onous exception
arise at specific locations during domain creatio= n and shutdown.
(Guillaume Munch-Maccagnoni, review by Gabriel Sc= herer)

- #11911, #12381: Restore statmemprof functionality in part, = with
some API changes in Gc.Memprof.
(Nick Barnes, revi= ew by Jacques-Henri Jourdan
   and Guillaume Munch-Maccagnoni)= .

- #12430: Simplify dynamic bytecode loading in Meta.reify_bytecode=
(Stephen Dolan, review by S=C3=A9bastien Hinderer, Vincent Lavir= on and Xavier
   Leroy)

- #12489: Fix an error-handling= bug in caml_alloc_sprintf
(Stephen Dolan, report by Chris Casing= hino, review by Jeremy Yallop
   and Xavier Leroy)

- #1= 1307: Finish adapting the implementation of asynchronous actions for
&nb= sp; multicore: soundness, liveness, and performance issues.
Do no= t crash if a signal handler is called from an unregistered C
thre= ad, and other possible soundness issues. Prevent issues where join
= ; on other domains could make the toplevel unresponsible to Ctrl-C. Avoid needless repeated polling in C code when callbacks cannot run
&= nbsp; immediately.
(Guillaume Munch-Maccagnoni, review by Enguerr= and Decorne, Xavier
Leroy, and KC Sivaramakrishnan)

- #126= 34: Simplify TSan backtrace bookkeeping upon raise
(Olivier Nicol= e and Fabrice Buoro, review by Gabriel Scherer)

* #12686: Some primi= tives had the wrong types to be callable from the bytecode
interp= reter. Either fix their types, mark them as `CAMLexport` instead of `CAMLprim`, or remove them entirely if no longer used.
(= Xavier Leroy, review by David Allsopp)

- #12700, continuing #11763 a= nd trying to address #12660:
Use the correct types for primitives= when generating the table of primitives
used by ocamlrun.
&nb= sp; (Xavier Leroy, motivation, review and improvements by Antonin D=C3=A9ci= mo)

- #12345, #12710: Fix issues with finaliser orphaning at domain = termination
(KC Sivaramakrishnan, report by Gabriel Scherer, revi= ew by Gabriel Scherer,
Sadiq Jaffer and Fabrice Buoro)

- #= 12599: Refactor Dynlink startup to avoid parsing bytecode sections twice (Stephen Dolan, review by David Allsopp, Hugo Heuzard, Damien Dolig= ez and
   Xavier Leroy)

- #12678, #12898: free channel = buffers on close rather than on finalization
(Damien Doligez, rev= iew by Jan Midtgaard and Gabriel Scherer, report
   by Jan Mid= tgaard)

- #12681: Fix TSan false positives due to volatile write han= dling
(Olivier Nicole, Fabrice Buoro and Anmol Sahoo, review by L= uc Maranget,
   Gabriel Scherer, Hernan Ponce de Leon and Xavi= er Leroy)

- #12743: Use pthread_sigmask instead of sigprocmask
&n= bsp; Updates usage of sigprocmask to pthread_sigmask in otherlibs/unix.
= (Max Slater, review by Miod Vallat and Xavier Leroy)

- #12769= : Unify MSVC and MinGW-w64 code paths, by always using WinAPI
dir= ectly.
(David Allsopp, Antonin D=C3=A9cimo, and Samuel Hym, revie= w by Nicolas
   Ojeda Bar)

- #11911, #12382, #12383: Re= store statmemprof functionality in part
(backtrace buffers, per-t= hread and per-domain data structures,
GC/allocation interface). (= Nick Barnes, review by Gabriel Scherer,
Fabrice Buoro, Sadiq Jaff= er, Guillaume Munch-Maccagnoni, and
Jacques-Henri Jourdan).
- #12735: Store both ends of the stack chain in continuations
(= Leo White, review by Miod Vallat and KC Sivaramakrishnan)

- #12746: = Simplify and clean up TSan annotations
(Olivier Nicole, review by= Miod Vallat and Fabrice Buoro)

- #12809: Add ThreadSanitizer suppor= t to FreeBSD/amd64
(Miod Vallat, review by Gabriel Scherer)
- #12810: Port ThreadSanitizer support to Linux and macOS on arm64
&nb= sp; (Miod Vallat, review by Tim McGilchrist)

- #12811: Define and us= e the CAMLthread_local macro for TLS variables.
(Antonin D=C3=A9c= imo and Samuel Hym, review by Miod Vallat and Xavier Leroy)

- #12814= : More detailed failure messages from `input_value` and `Marshal.from_*` (Xavier Leroy, review by Stephen Dolan and Anil Madhavapeddy)
- #12815: Correctly format multi-line locations in exception backtraces, = in the
style that the compiler driver uses.
(David Alls= opp, review by Gabriel Scherer)

- #12773, #12830, #12834: Rewrite `c= aml_c_thread_(un)register` to fix
various bugs.
(Guilla= ume Munch-Maccagnoni, reported by Miod Vallat, suggested by
  = Hari Hara Naveen S, reviewed by Fabrice Buoro, Gabriel Scherer and
&nbs= p; Miod Vallat)

- #12876: Port ThreadSanitizer support to Linu= x on POWER
(Miod Vallat, review by Tim McGilchrist)

- #128= 86: Reinitialize IO mutexes after fork
(Max Slater, review by Gui= llaume Munch-Maccagnoni and Xavier Leroy)

- #12907: Port ThreadSanit= izer support to Linux on RiscV
(Miod Vallat, review by Nicol=C3= =A1s Ojeda B=C3=A4r and Fabrice Buoro)

- #12915: Port ThreadSanitize= r support to Linux on s390x
(Miod Vallat, review by Tim McGilchri= st)

- #12934: Fix data races between marking and sweeping functions<= br> (Olivier Nicole, suggested by Stephen Dolan, review by Gabriel Sc= herer,
   Miod Vallat and Damien Doligez)

### Language = features:

- #12295, #12568: Give `while true' a polymorphic type, si= milarly to
`assert false'
(Jeremy Yallop, review by Nic= ol=C3=A1s Ojeda B=C3=A4r and Gabriel Scherer,
suggestion by Rodol= phe Lepigre and John Whitington)

- #12044: Add local module open syn= tax for types.
```
    module A =3D struct
&= nbsp;     type t =3D int
    &nb= sp; type r =3D unit
      type s =3D string
=     end

    type example =3D A.(t * r = * s)
```
(Alistair O'Brien, review by Gabriel Scherer, = Nicol=C3=A1s Ojeda B=C3=A4r
   and Florian Angeletti)

-= #11252, RFC 27: Support raw identifier syntax \#foo
(Stephen Dol= an, review by David Allsopp, Gabriel Scherer and Olivier Nicole)

- #12315: Use type annotations from arguments in let rec
(Stephe= n Dolan, review by Gabriel Scherer)

- #12375: allow use of [@untagge= d] for all immediate types like char, bool,
and variant with only= constant constructors.
(Christophe Raffalli, review by Gabriel S= cherer)

* #12502: the compiler now normalizes the newline sequence \= r\n to
a single \n character during lexing, to guarantee that the= semantics
of newlines in string literals is not modified by Wind= ows tools
transforming \n into \r\n in source files.
Wa= rning 29 [eol-in-string] is not emitted anymore, as the normalization
&n= bsp; gives a more robust semantics to newlines in string literals.
= ; (Gabriel Scherer and Damien Doligez, review by Daniel B=C3=BCnzli, David<= br>   Allsopp, Andreas Rossberg, Xavier Leroy, report by Andreas = Rossberg)

### Standard library:

- #11563: Add the Dynarray mo= dule to the stdlib. Dynamic arrays are
arrays whose length can be= changed by adding or removing elements at
the end, similar to 'v= ectors' in C++ or Rust.
(Gabriel Scherer, Simon Cruanes and Flori= an Angeletti, review by
   Daniel B=C3=BCnzli, Guillaume Munch= -Maccagnoni, Cl=C3=A9ment Allain,
   Damien Doligez, Wiktor Ku= chta and Pieter Goetschalckx)

- #12716: Add `Format.pp_print_not= hing` function.
(L=C3=A9o Andr=C3=A8s, review by Gabriel Scherer = and Nicol=C3=A1s Ojeda B=C3=A4r)

* #6732, #12423: Make Buffer.add_su= bstitute surjective and fix its
documentation.
(Damien = Doligez, review by Antonin D=C3=A9cimo)

* #10775, #12499: Half-preci= sion floating-point elements in Bigarray.
(Anton Yabchinskiy, rev= iew by Xavier Leroy and Nicol=C3=A1s Ojeda B=C3=A4r)

- #11517, #1247= 7: Expose pp_infinity in interface of the format module, and
chec= k that margin is less than pp_infinity when setting or checking geometry. (Janith Petangoda, reported by Simmo Saan, reviewed by Florian Ang= eletti,
Simmo Saan, Josh Berdine and Gabriel Scherer)

- #1= 2217: Add `Array.shuffle`.
(Daniel B=C3=BCnzli, review by Nicol= =C3=A1s Ojeda B=C3=A4r, David Allsopp and Alain Frisch)

- #12212: Ad= d cache-aligned constructor for atomics. The patch ensures that
a= ll allocations (of the right size) in the shared heap are aligned.
= ; (Bartosz Modelski with Gabriel Scherer, Guillaume Munch-Maccagnoni,
&n= bsp; Xavier Leroy, review by Alain Frisch, Anil Madhavapeddy, Gabriel= Scherer,
   Guillaume Munch-Maccagnoni, KC Sivaramakrishnan, = Stefan Muenzel,
   Xavier Leroy)

- #12307: Add BLAKE2b = hashing and an MD5 submodule to the Digest module.
(Xavier Leroy,= review by Olivier Nicole, Gabriel Scherer, Wiktor Kuchta,
   = Daniel B=C3=BCnzli, David Allsopp)

- #12365: Add In_channel.input_bi= garray, In_channel.really_input_bigarray,
Out_channel.output_biga= rray, Unix.read_bigarray, Unix.write_bigarray,
Unix.single_write_= bigarray.
(Nicol=C3=A1s Ojeda B=C3=A4r, review by Jeremy Yallop, = Xavier Leroy, Gabriel Scherer,
   David Allsopp)

- #124= 55: Add `Array.init_matrix`, `Float.Array.make_matrix`,
`Float.Ar= ray.init_matrix`.
(Glen M=C3=A9vel, review by Xavier Leroy, Gabri= el Scherer, Jeremy Yallop,
Nicolas Ojeda Bar)

* #12455: `A= rray.make_matrix dimx dimy f` now raises `Invalid_argument`
when = `dimx =3D 0 && dimy < 0` This was already specified but not enfo= rced.
(Glen M=C3=A9vel, report by Jeremy Yallop, review by Nicola= s Ojeda Bar)

- #12459: Add `Random.int_in_range`, `Random.int32_in_r= ange`,
`Random.int64_in_range`, `Random.nativeint_in_range`,
&= nbsp; and their counterpart in `Random.State`.
(Glen M=C3=A9vel a= nd Xavier Leroy, review by Gabriel Scherer, Xavier Leroy,
   F= lorian Angeletti)

- #12459: `Random`: restore compatibility between = 32-bit integers (JavaScript)
and 63-bit integers (64-bit OCaml).<= br> For `Random.full_int` this was guaranteed in 4.14 but wrongly rem= oved in 5.0.
(Xavier Leroy, review by Glen M=C3=A9vel)

- #= 12511: Minor performance improvements and cleanups in the implementation of modules Int32, Int64, and Nativeint
(Xavier Leroy, rev= iew by Gabriel Scherer and Daniel B=C3=BCnzli)

- #12558: Adapt GC al= arms for multicore and fix their documentation.
(Guillaume Munch-= Maccagnoni, review by KC Sivaramakrishnan
   and Gabriel Scher= er)

- #12625: Remove the Closure module from Obj
(Vincent = Laviron, review by Xavier Leroy)

- #12758, #12998: Remove the `Marsh= al.Compression` flag to the
`Marshal.to_*` functions. The c= ompilers are still able to use
ZSTD compression for compilation a= rtefacts.
This is a forward port and clean-up of the emergency fi= x that was introduced
in OCaml 5.1.1 by #12734.
(Xavier= Leroy, review by Damien Doligez)

- #12784: Fix computation of minor= -heap allocation in Gc.counters()
and Gc.allocated_bytes(). (Nick= Barnes, review by Gabriel Scherer)

- #12770: Add `Fun.compose`.
= (Justin Frank, review by Nicol=C3=A1s Ojeda B=C3=A4r, Daniel B=C3=BC= nzli and Jeremy Yallop)

- #12845: Add `{In,Out}_channel.is_binary_mo= de` as the dual of
`set_binary_mode`. This function was previousl= y only available in the internal
C API.
(David Allsopp,= review by Nicol=C3=A1s Ojeda B=C3=A4r and Xavier Leroy)

### Type sy= stem:

- #12313, #11799: Do not re-build as-pattern type when a groun= d type annotation
is given. This allows to work around problems w= ith GADTs in as-patterns.
(Jacques Garrigue, report by Leo White,= review by Gabriel Scherer)

### Code generation and optimizations:
- #11239: on x86-64 and RISC-V, reduce alignment of OCaml stacks from= 16 to 8.
This reduces stack usage. It's only C stacks that= require 16-alignment.
(Xavier Leroy, review by Gabriel Scherer a= nd Stephen Dolan)

- #12311: on POWER, 32-bit FP numbers stored in me= mory (e.g. in bigarrays)
were not correctly rounded sometimes. (Xavier Leroy, review by Anil Madhavapeddy and Tim McGilchrist)
=
- #12551, #12608, #12782, #12596: Overhaul of recursive value compilati= on.
Non-function recursive bindings are now forbidden from Lambda= onwards,
and compiled using a new Value_rec_compiler module.
= (Vincent Laviron and Lunia Ayanides, review by Gabriel Scherer,
&= nbsp; Stefan Muenzel and Nathana=C3=ABlle Courant)

- #1809, #1= 2181: rewrite `compare x y op 0` to `x op y` when values are integers
&n= bsp; (Xavier Clerc, Stefan Muenzel, review by Gabriel Scherer and Vincent L= aviron)

- #12825: disable common subexpression elimination for atomi= c loads... again.
(Gabriel Scherer, review by KC Sivaramakrishnan= , Xavier Leroy
   and Vincent Laviron, report by Vesa Karvonen= )

### Other libraries:

- #12213: Dynlink library, improve leg= ibility of error messages
(Samuel Hym, review by Gabriel Scherer = and Nicol=C3=A1s Ojeda B=C3=A4r)

* #12686: Runtime_events library, C= API: define
`caml_runtime_events_{start,pause,resume}` as return= ing `void`
instead of `value`.
(Xavier Leroy, review by= David Allsopp)

### Compiler user-interface and warnings:

- #= 11989, #12246, RFC 31: New flag, -H, to allow for transitive dependencies without including them in the initial environment.
(Chri= s Casinghino, Fran=C3=A7ois Bobot, and Gabriel Scherer, review by Leo White=
   and Stefan Muenzel, RFC by Fran=C3=A7ois Bobot)

* #10613, #12405: Simplify the values used for the system variable (`syste= m:` in
`ocamlopt -config` or the `Config.system` constant). In pa= rticular, s390x and
ppc64 now report "linux" instead of "elf"; al= l variants of 32-bit ARM on Linux
now report "linux"; OpenBSD now= reports "openbsd" instead of "bsd" for 32-bit
ARM; FreeBSD, NetB= SD and OpenBSD now report the same value for both x86_64 and
x86_= 32; x86_32 systems matching *bsd but not freebsd*, netbsd* or openbsd*
&= nbsp; are no longer identified (as on x86_64); x86_32 Linux now reports "li= nux"
instead of "linux_elf".
(David Allsopp, request by= Kate Deplaix, review by S=C3=A9bastien Hinderer and
Xavier Leroy= )

- #12247: configure: --disable-ocamldebug can now be used instead<= br> of --disable-debugger (which remains available for compatibility)=
(Gabriel Scherer, review by Damien Doligez and S=C3=A9bastien Hi= nderer)

- #12199: improve the error message for non-overriding `inhe= rit!`
(Florian Angeletti, review by Jules Aguillon)

- #122= 10: uniform style for inline code in compiler messages
(Florian A= ngeletti, review by Gabriel Scherer)

* #12278, #:12325: Remove the O= CAML_FLEXLINK environment variable from the
compiler drivers. Thi= s environment variable was previously used as part of the
FlexDLL= bootstrap procedure and existed solely for that purpose. Its removal
&n= bsp; greatly simplifies both the build system and testsuite machinery.
&= nbsp; (David Allsopp, review by S=C3=A9bastien Hinderer)

- #12347: e= rror messages: always report missing polyvariant tags
(Florian An= geletti, report by Tianbo Hao, review by Gabriel Scherer)

- #12224, = specialized error message when trying to apply non-functor
module= (e.g `module M =3D Int(Int)`)
(Florian Angeletti, review by Gabr= iel Scherer)

- #12451: Warning 53 (misplaced attributes) now works f= or all attributes.
(Chris Casinghino, review by Florian Angeletti= )

- #12622: Give hints about existential types appearing in error me= ssages
(Leo White, review by Gabriel Scherer and Florian Angelett= i)

- #12671: When a class type parameter or class parameter does not= match,
identify which parameter in the error message, instead of= saying
"A type parameter" or "A parameter".
(Stefan Mu= enzel, review by Gabriel Scherer)

- #12679: Add more detail to the e= rror message and manual in case of
invalid module type substituti= ons.
(Stefan Muenzel, review by Gabriel Scherer and Florian Angel= etti)

- #12750: Display the command executed to extract primitives i= n
`ocamlc -verbose`.
(David Allsopp, review by Nicol=C3= =A1s Ojeda B=C3=A4r)

- #12777: Add details about the actual and expe= cted method types to the method
mismatch error messages.
= ; (Javier Ch=C3=A1varri, review by Gabriel Scherer and Florian Angeletti)
* #12942: Fix an line ordering in some module inclusion error message= s
(Nick Roberts, review by Florian Angeletti, report by Carl East= lund)
### Manual and documentation:

- #12338: clarification of th= e documentation of process related function in
the unix module re= garding the first element of args and shell's pid.
(Christophe Ra= ffalli, review by Florian Angeletti)

- #12473: Document in runtime/m= emory.c our current understanding of
accesses to the OCaml heap f= rom the C runtime code -- the problem
of hybrid programs mixing t= wo memory models.
(Gabriel Scherer and Guillaume Munch-Maccagnoni= , review by Olivier
   Nicole and Xavier Leroy)

- #1245= 6: Document the incompatibility between effects on the one
hand, = and `caml_callback` and asynchronous callbacks (signal
handlers, = finalisers, memprof callbacks...) on the other hand.
(Guillaume M= unch-Maccagnoni, review by KC Sivaramakrishnan)

- #12694: Document i= n runtime/tsan.c the TSan instrumentation choices and the
consequ= ences with regard to the memory model.
(Olivier Nicole, review by= Miod Vallat, Gabriel Scherer, Guillaume
   Munch-Maccagnoni a= nd Fabrice Buoro)

- #12802: Add manual chapter about ThreadSanitizer= support
(Olivier Nicole, review by Miod Vallat, Sebastien Hinder= er, Fabrice Buoro,
   Gabriel Scherer and KC Sivaramakrishnan)=

- #12819: Clarify which runtime interactions are allowed in custom = ops
(Basile Cl=C3=A9ment, review by Guillaume Munch-Maccagnoni an= d Xavier Leroy)

- #12840: manual: update runtime tracing chapter for= custom events (ex #12335)
(Lucas Pluvinage, Sadiq Jaffer and Oli= vier Nicole, review by Gabriel Scherer,
   David Allsopp, Tim = McGilchrist and Thomas Leonard)

- #13066, update OCAMLRUNPARAM docum= entation for the stack size parameter l
(Florian Angeletti, revie= w by Nicol=C3=A1s Ojeda B=C3=A4r, Tim McGilchrist, and
   Miod= Vallat)

- #13078: update Format tutorial on structural boxes to men= tion alignment
questions.
(Edwin T=C3=B6r=C3=B6k, revie= w by Florian Angeletti)

- #13092: document the existence of the `[@@= poll error]` built-in attribute
(Florian Angeletti, review by Gab= riel Scherer)

### Tools:

- #12340: testsuite: collect known i= ssues with current -short-paths
implementation for existential ty= pes
(Florian Angeletti, Samuel Hym, review by Florian Angeletti a= nd Thomas Refis)

- #12147: ocamllex: Allow carriage returns at the e= nd of line directives.
(SeungCheol Jung, review by Nicol=C3=A1s O= jeda B=C3=A4r)

- #12260: Fix invalid_argument on some external or mo= dule aliases in ocamlnat
(Fabian Hemmer, review by Vincent Laviro= n)

- #12185: New script language for ocamltest.
(Damien Do= ligez with Florian Angeletti, S=C3=A9bastien Hinderer, Gabriel Scherer,
=    review by S=C3=A9bastien Hinderer and Gabriel Scherer)

= - #12371: ocamltest: fix recursive expansion of variables.
(Anton= in D=C3=A9cimo, Damien Doligez, review by S=C3=A9bastien Hinderer,
= ; Damien Doligez, Gabriel Scherer, and Xavier Leroy)

* #12497,= #12613: Make ocamlc/ocamlopt fail with an error when no
input fi= les are specified to build an executable.
(Antonin D=C3=A9cimo, r= eview by S=C3=A9bastien Hinderer)

- #12576: ocamldep: various refact= ors.
(Antonin D=C3=A9cimo, review by Florian Angeletti, Gabriel S= cherer, and L=C3=A9o Andr=C3=A8s)

- #12615: ocamldoc: get rid of the= odoc_literate and odoc_todo generators.
(S=C3=A9baistien Hindere= r, review by Gabriel Scherer and Florian Angeletti)

- #12624: Use $X= DG_CONFIG_DIRS in addition to $XDG_CONFIG_HOME when searching
for= init.ml and use this to extend init.ml support to the toplevel when
&nb= sp; running on Windows.
(David Allsopp, report by Jonah Beckford,= review by Nicol=C3=A1s Ojeda B=C3=A4r and
   Antonin D=C3=A9c= imo)

- #12688: Setting the env variable `NO_COLOR` with an empty val= ue no longer
has effects. Previously, setting `NO_COLOR` with any= value, including
the empty value, would disable colors (unless `= OCAML_COLOR` is also set).
After this change, the user must set `= NO_COLOR` with an non-empty value
to disable colors. This reflect= s a specification clarification/change
from the upstream website = at https://no-color.org.
(Favonia, review by Gabriel Scherer)
=
- #12744: ocamltest: run tests in recursive subdirs more eagerly
&nb= sp; (Nick Roberts, review by Nicol=C3=A1s Ojeda B=C3=A4r)

- #12901, = 12908: ocamllex: add overflow checks to prevent generating incorrect
&nb= sp; lexers; use unsigned numbers in the table encoding when possible.
&n= bsp; (Vincent Laviron, report by Edwin T=C3=B6r=C3=B6k, review by Xavier Le= roy)

### Internal/compiler-libs changes:

- #12508 : Add compi= ler-side support for project-wide occurrences in Merlin, by
gener= ating index tables of all identifier occurrences. This extra data in .cmt files is only added when the new flag -bin-annot-occurrences is pa= ssed.
(Ulysse G=C3=A9rard, Nathana=C3=ABlle Courant, suggestions = by Gabriel Scherer and Thomas
Refis, review by Florian Angeletti,= Gabriel Scherer and Thomas Refis)

- #12236, #12386, #12391, #12496,= #12673: Use syntax as sole determiner of arity
This changes func= tion arity to be based solely on the source program's
parsetree. = Previously, the heuristic for arity had more subtle heuristics
th= at involved type information about patterns. Function arity is import= ant
because it determines when a pattern match's effects run and = is an input
into the fast path for function application.

&= nbsp; This change affects tooling: it changes the function constructs in pa= rsetree
and typedtree.

See https://github.com/ocaml= /RFCs/pull/32 for the original RFC.

(Nick Roberts; review by = Richard Eisenberg, Leo White, and Gabriel Scherer;
RFC by Stephen= Dolan)

- #12639: parsing: Attach a location to the RHS of Ptyp_= alias
and improve the 'alias type mismatch' error message.
&nb= sp; (Jules Aguillon, review by Florian Angeletti)

- #12447: Remove 3= 2-bit targets from X86_proc.system
(Masanori Ogino, review by Dav= id Allsopp)

- #12216, #12248: Prevent reordering of atomic loads dur= ing instruction
scheduling. This is for reference, as instr= uction scheduling is currently
unused in OCaml 5.
(Xavi= er Leroy, report by Luc Maranget and KC Sivaramakrishnan,
   r= eview by Nicol=C3=A1s Ojeda B=C3=A4r)

- #12025: Split Typecore.unify= _pat_types into two
to avoid unnecessary references to the enviro= nment in type_pat
(Jacques Garrigue and Takafumi Saikawa, review = by Gabriel Scherer)

- #12031: Use dedicated types to represent names= of compilation units
and predefined exceptions in CMO files.
= (S=C3=A9bastien Hinderer, review by Florian Angeletti, Thomas Refis,=
Gabriel Scherer, Vincent Laviron, Pierre Chambart, Luke Maurer,<= br> Hugo Heuzard, Xavier Leroy and Damien Doligez)

- #12109: P= ack parameters to unification in unification_environment
(Takafum= i Saikawa and Jacques Garrigue, review by Richard Eisenberg)

- #1233= 1, #12361: Pack the unification data for pattern checking in Typecore
&n= bsp; (Takafumi Saikawa and Jacques Garrigue,
   review by Gabr= iel Scherer, Thomas Refis and Florian Angeletti)

- #12229: Remove gl= obal mutable state for typechecking patterns
in Typecore in favor= of local mutable state.
(Nick Roberts, review by Takafumi Saikaw= a)

- #12542: Minor bugfix to #12236: restore dropped call to `instan= ce`
(Nick Roberts, review by Jacques Garrigue)

- #12242: M= ove the computation of stack frame parameters to a separate
`Stac= kframe` module, and save the parameters in the results of the
`Li= nearize` pass
(Xavier Leroy, review by KC Sivaramakrishnan and Ma= rk Shinwell)

- #12442: document jump summaries in the pattern-matchi= ng compiler
(Gabriel Scherer and Thomas Refis, review by Florian = Angeletti
   and Vincent Laviron)

- #12446, #12792: rem= ove the hooks machinery around channel locking
in runtime/io.c (Gabriel Scherer, review by Xavier Leroy)

- #12389, #12544, = #12984, #12987: centralize the handling of metadata for
compilati= on units and artifacts in preparation for better unicode support for
&nb= sp; OCaml source files.
(Florian Angeletti, review by Vincent Lav= iron and Gabriel Scherer)

- #12532, #12553: improve readability of t= he pattern-matching debug output
(Gabriel Scherer, review by Thom= as Refis)

- #12537: Use C11/C++11 standard static assertion.
&nbs= p; (Antonin D=C3=A9cimo, review by Sebastien Hinderer, Xavier Leroy,
&nb= sp; and KC Sivaramakrishnan)

- #12169: runtime: document and e= nforce naming conventions around STW sections.
(Gabriel Scherer, = review by Enguerrand Decorne, Miod Vallat, B. Szilvasy
   and = Nick Barnes, report by KC Sivaramakrishnan)

- #12669 : Clean up some= global state handling in schedgen
(Stefan Muenzel, review by Mio= d Vallat and Gabriel Scherer)

- #12640: Make the module separator us= ed in symbol names configurable
(Miod Vallat, review by Hugo Heuz= ard and Xavier Leroy)

- #12691 : Clean up Ctype.expand_abbrev_gen an= d
rename Env.add_local_type to add_local_constraint
(Ta= kafumi Saikawa and Jacques Garrigue, review by Florian Angeletti)

- = #12786 : Clean up the algorithm of Ctype.limited_generalize
(Taka= fumi Saikawa and Jacques Garrigue, review by Gabriel Scherer)

- #106= 91: Ast_mapper, Ast_iterator: add functions directive_argument,
t= oplevel_directive and toplevel_phrase.
(Guillaume Petiot, review = by Gabriel Scherer and Kate Deplaix)

- #12764: Move all installable = headers in `caml/` sub-directories.
(Antonin D=C3=A9cimo, review = by Gabriel Scherer and David Allsopp)

- #12914: Slightly change the = s390x assembly dialect in order to build with
Clang's integrated = assembler.
(Miod Vallat, review by Gabriel Scherer)

- #130= 01: do not read_back entire shapes to get aliases' uids when building the usages index
(Ulysse G=C3=A9rard, review by Gabriel Sche= rer and Nathana=C3=ABlle Courant)

### Build system:

- #12198,= #12321, #12586, #12616, #12706, #13048: continue the merge of the
= ; sub-makefiles into the root Makefile started with #11243, #11248,
&nbs= p; #11268, #11420 and #11675.
(S=C3=A9bastien Hinderer, review by= David Allsopp and Florian Angeletti)

- #12569, #12570: remove 'othe= rlibraries' as a prerequisite for 'runtop';
use 'runtop-with-othe= rlibs' to use a library from otherlibs/
(Gabriel Scherer, review = by S=C3=A9bastien Hinderer, suggestion by David Allsopp)

- #12652: M= ake magic numbers easier to bump and duplicate.
(S=C3=A9bastien H= inderer, review by Antonin D=C3=A9cimo, David Allsopp and Florian
= Angeletti)

* #12751: --with-target-bindir configure option implemen= ted. This option
refers to the location of the runtime binaries o= n the target system for a
cross-compiler and is embedded in execu= tables produced by ocamlc. It does
not affect the bytecode execut= ables installed as part of the build. The old
mechanism `make TAR= GET_BINDIR=3D..` no longer works.
(David Allsopp, review by Damie= n Doligez, Xavier Leroy and Olivier Nicole)

- #12768, #13030: Detect= mingw-w64 coupling with GCC or LLVM, detect clang-cl,
and fix C = compiler feature detection on macOS.
(Antonin D=C3=A9cimo, review= by Miod Vallat and S=C3=A9bastien Hinderer)

### Bug fixes:

-= #10652, #12720: fix evaluation order in presence of optional arguments
= (Jacques Garrigue, report by Leo White, review by Vincent Laviron)
- #12595, #12597: fix a race in `caml_clear_gc_stats_sample`
= ; (Gabriel Scherer, review by B. Szilvasy, report by B. Szilvasy)

- = #12580: Fix location of alias pattern variables.
(Chris Casinghin= o, review Gabriel Scherer, report by Milo Davis)

- #12583: Add a clo= sing event for when `EV_MAJOR_EPHE_MARK` is complete
(Sudha Parim= ala, review by Gabriel Scherer)

- #12566: caml_output_value_to_mallo= c wrongly uses `caml_stat_alloc`
instead of `malloc` since 4.06, = breaking (in pooled mode) user code
that uses `free` on the resul= t. Symmetrically,
caml_input_value_from_malloc should use `free`.=
(Gabriel Scherer, review by Xavier Leroy and Enguerrand Decorne,=
   report by Ido Yariv)

- #12490: Unix: protect the po= pen_processes hashtable with a mutex
(Gabriel Scherer, report by = Olivier Nicole, review by Xavier Leroy)

- #11931: Fix tricky typing = bug with type substitutions
(Stephen Dolan, review by Leo White a= nd Jacques Garrigue)

- #12037, #12171: Fix get_extern_state potentia= l NULL dereference.
(Alexander Skvortsov, report by T=C3=B6r=C3= =B6k Edwin,
   design by Gabriel Scherer, Xavier Leroy)
- #12635: Fix get_intern_state potential NULL dereference.
(Anto= nin D=C3=A9cimo, review by KC Sivaramakrishnan)

- #12032, #12059: Bu= g fixes related to compilation of recursive definitions
(Vincent = Laviron, report by Victoire Noizet, review by Gabriel Scherer)

* #12= 145: Loopy constraints cause ocamlc to loop.
Fixed by completely = removing the call to `update_type` in
`Typedecl.transl_type_decl`= , as the expansion is already checked by
`check_regularity`. As a= result, recursion is more polymorphic,
which may cause some (ess= entially wrong) type declarations to have
unbound type variables,= and some constraints unrelated to the concrete
type to be ignore= d (see tests/typing-misc/constraints.ml).
(Jacques Garrigue, repo= rt by Richard Eisenberg, review by Leo White)

- #12207, #12222: Make= closure computation linear in the number of recursive
functions = instead of quadratic
(Vincent Laviron, report by Fran=C3=A7ois Po= ttier, review by Nathana=C3=ABlle Courant
and Gabriel Scherer)
- #11040, #12591: fix a data race in major_gc.c
(Gabriel Sch= erer, review by Guillaume Munch-Maccagnoni
   and KC Sivaramak= rishnan, report by Sadiq Jaffer)

- #12238, #12403, #12698: read inpu= t files in one go to avoid source reprinting
issues.
(G= abriel Scherer, report by Mike Spivey and Vincent Laviron, review by
&nb= sp; Nicol=C3=A1s Ojeda B=C3=A4r, Xavier Leroy and Jeremy Yallop)
<= br>- #12334, #12368: Bad error message with mutually recursive abbreviation= s
(Jacques Garrigue, report by Richard Eisenberg, review by Gabri= el Scherer
   and Richard Eisenberg)

- #12401: `seek_in= ` and `seek_out` sometimes returned normally when given
negative = offsets, instead of failing. Now both functions should consistently
&nbs= p; raise `Sys_error` in this case.
(Nicol=C3=A1s Ojeda B=C3=A4r, = review by Gabriel Scherer)

- #12267: Fix stack alignment computation=
(Miod Vallat, report by Jan Midtgaard, review by Gabriel Scherer= )

- #12395, #12404: Fix thread-unsafety in the fallback implementati= on of
`Unix.create_process` (the one used when `posix_spawnp` is = unavailable)
(Xavier Leroy, report by Chris Vine, review by Nicol= =C3=A1s Ojeda B=C3=A4r)

- #12949: open shadowing mistriggers
&nbs= p; (Gabriel Scherer, review by Florian Angeletti, report by Andreas Rossber= g)

- #12526: Honor `ocaml.inline always` attribute on functions with=
optional arguments and default values in the Closure backend
= (Alain Frisch, review by Vincent Laviron)

- #12677, #12889: m= ake Domain.DLS thread-safe
(Gabriel Scherer, review by Olivier Ni= cole and Damien Doligez,
   report by Vesa Karvonen)

- = #12561: Fix crash when combining TSan and frame-pointers
(Fabrice= Buoro and Olivier Nicole, report by Jan Midtgaard, review by Miod
= ; Vallat and Gabriel Scherer)

- #12482: Rework bounds checking code = in the POWER backend
(Miod Vallat and Xavier Leroy, report by Jan= Midtgaard, review by
   KC Sivaramakrishnan)

- #12528,= #12703: Avoid pointer arithmetic overflow in Tag_val macro
(very= likely harmless, but can trigger alarms)
(Xavier Leroy, report b= y Sam Goldman, review by Guillaume Munch-Maccagnoni)

- #12593: TSan = should handle Effect.Unhandled correctly
(Fabrice Buoro and Olivi= er Nicole, report by Jan Midtgaard and Miod Vallat,
   review = by Gabriel Scherer)

- #12684: fix locations filename in AST produced= by the `-pp` option
(Gabriel Scherer, review by Florian Angelett= i)

- #12714: check whether macros are defined before using them to e= nsure
that the headers can always be used in code which turns on = -Wundef
(or equivalent).
(Antonin D=C3=A9cimo, review b= y Miod Vallat, Gabriel Scherer,
   Xavier Leroy, and David All= sopp)

- #12726: fix segmentation fault under Windows when executing = a bytecode file if
the runtime (`ocamlrun.exe`) cannot be found.<= br> (Vadim Zborovskii, Nicol=C3=A1s Ojeda B=C3=A4r, report by Vadim Z= borovskii, review by
David Allsopp)

- #12727, #12730: fix = bug with value let-rec and labelled applications
(Vincent Laviron= , review by Gabriel Scherer)

* #12751: Always keep within the 128 ch= aracter limit for shebang lines. Previous
fix in #8622 handled bu= ilding the compiler with a long prefix; this patch
extends this t= o the bytecode executables emitted by that compiler.
(David Allso= pp, review by Damien Doligez, Xavier Leroy and Olivier Nicole)

- #12= 755: Fix data race on global pools arrays of pool_freelist
(Fabri= ce Buoro and Olivier Nicole, review by Gabriel Scherer)

- #12796, #1= 2801: Fix memory corruption in caml_unix_alloc_sockaddr
(Thomas L= eonard, review by Nicol=C3=A1s Ojeda B=C3=A4r)

- #12737: Fix data ra= ces in minor_gc.c and caml_natdynlink_open
(Olivier Nicole, revie= w by Stefan Muenzel, Miod Vallat, Guillaume
   Munch-Maccagnon= i, Gabriel Scherer and Xavier Leroy)

- #12831: Fix call to caml_call= _realloc_stack for s390x in PIC mode
(Vincent Laviron, report by = Jerry James, review by Miod Vallat)

* #12837: Show non-generalizable= type parameters in type definitions
Changes type of type paramet= ers in outcometree.mli.
(Jacques Garrigue, review by Richard Eise= nberg)

- #12897: fix locking bugs in Runtime_events
(Gabri= el Scherer and Thomas Leonard,
   review by Olivier Nicole, Vi= ncent Laviron and Damien Doligez,
   report by Thomas Leonard)=

- #12851: Fix race between runtime events teardown and event emissi= on
(Olivier Nicole, review by Miod Vallat and Gabriel Scherer)
- #12860: Fix an assertion that wasn't taking into account the possibi= lity of an
ephemeron pointing at static data.
(Mark Shi= nwell, review by Gabriel Scherer and KC Sivaramakrishnan)

- #12861: = Fix a possible crash in the `threads` library.
(Mark Shinwell, re= view by Gabriel Scherer and KC Sivaramakrishnan)

- #11040, #12894: S= ilence false data race observed between caml_shared_try_alloc
and= oldify. Introduces macros to call tsan annotations which help annotate
= a ``happens before'' relationship.
(Hari Hara Naveen S and= Olivier Nicole,
   review by Gabriel Scherer and Miod Vallat)=

- #12958: Fix tail-modulo-cons compilation of try-with, && = and ||
expressions.
(Gabriel Scherer and Nicol=C3=A1s O= jeda B=C3=A4r, report by Sylvain Boilard, review by
Gabriel Scher= er)

- #12919: Fix register corruption in caml_callback2_asm on s390x= .
(Miod Vallat, review by Gabriel Scherer)

- #12924, #1293= 0: Rework package constraint checking to improve interaction with
= immediacy
(Chris Casinghino and Florian Angeletti, review by Flo= rian Angeletti and
   Richard Eisenberg)

- #12969: Fix = a data race in caml_darken_cont
(Fabrice Buoro and Olivier Nicole= , review by Gabriel Scherer and Miod Vallat)

- #12971, #12974: fix a= n uncaught Ctype.Escape exception on some
invalid programs formin= g recursive types.
(Gabriel Scherer, review by Florian Angeletti,= report by Neven Villani)

- #13019: Remove linking instructions for = the Unix library from threads.cma
(this was done for threads.cmxa= in OCaml 3.11). Eliminates warnings from
new lld when using thre= ads.cma of duplicated libraries.
(David Allsopp, review by Nicol= =C3=A1s Ojeda B=C3=A4r)

- #13058: Add TSan instrumentation to caml_c= all_gc(), since it may raise
exceptions.
(Fabrice Buoro= , Olivier Nicole, Gabriel Scherer and Miod Vallat)

- #13079: Save an= d restore frame pointer across Iextcall on ARM64
(Tim McGilchrist= , review by KC Sivaramakrishnan and Miod Vallat)

- #13094: Fix undef= ined behavior of left-shifting a negative number.
(Antonin D=C3= =A9cimo, review by Miod Vallat and Nicol=C3=A1s Ojeda B=C3=A4r)

- #1= 3130: minor fixes to pprintast for raw identifiers and local module open syntax for types.
(Chet Murthy, review by Gabriel Scherer= )

--=_78db1b8c-e0a0-4bea-98e2-9be53ec9c931--