Merge tools/Makefile into the root Makefile

[shindere]

This continues the work started in #11243, #11248, #11268 and #11420.

As these former pull requests, the present one is best reviewed in a
commit-by-commit way.

The PR starts by fixing the build of ocamllex by taking the link flags
into account again -- they got accidentally lost in #11420.

After a few minor commits, a more substantial one adjusts the compilation
flags used to compile and link the programs in the tools/ directory so that
they match those used by the root Makefile. In particular, this
implied the creation of a few interface files which had been forgotten
in #11288, which in a way demonstrates how hard it was/is to e.g. enable
or disable a given warning all over our codebase.

In the same vein, a few commits work on distinguishing the compilation
and link stages for a few tools: rather than building them from their .ml
sources directly, we make sure to first compile them to .cmo files and then
only link them. This is so that the compilation can use the already existing
compilation rules rather than dedicated ones and, that way, make sure
the same flags are passed to the compilers consistently.

This PR also starts to use GNU make's VPATH variable to let make
know where to look for prerequisites. This variable is then
used to compute which directories to pass to the compiler as arguments
of -I. That way, we make sure the compiler and GNU make agree
on where to look for files.

In addition to a few other simplifications, among which there is the
computation of the LN variable during configure rather than during build,
this PR extends the existing build framework with a few build
variables and, more importantly, with a few macros that make linking OCaml
programs more concise and uniform.

Basically, to get a program foo linked, list its libraries in
foo_LIBRARIES, its modules in foo_MODULES and call one of the provided
link macros depending on whether you want to produce both bytecode and
native versions of foo (with OCAML_PROGRAM), only its bytecode
version (with OCAML_BYTECODE_PROGRAM) or only its native versoin
(with, guess?... OCAML_NATIVE_PROGRAM, yes).

Finally, before this PR the cmt and cmit source artifacts of
tools/ocamlmktop_init were not installed as they should have been. The PR
fixes this and installs them.

[dra27]

As ever, a thorough piece of work, and quite the nettle you've grasped again with this one, @shindere! It's pleasing that a 353 line Makefile is removed with only 209 lines added elsewhere in its place :)

I was concerned - incorrectly - that this might affect Cygwin64 when built in bootstrapped flexdll mode but this is working fine (I think that's due to a combination of your work on utils/Makefile and some of mine on $(MKEXE) et al). I also bumped all the magic numbers, and bootstrapping also seems to be working as expected.

I think you could squash the "Makefile.common: simplify the generic macros" into the previous commit, but it's fine if you want to keep the two separate versions.

Most of my comments are cosmetic or typos - the only major thing which needs fixing is the doubling of commands caused by ocamltoolsopt and ocamltoolsopt.opt both making the same recursive call to make in parallel in the opt.opt target.

In passing:

• In tools/ocamlmklib.ml, the syslib function is a leftover from ocamlbuild days and the chop_suffix functions appears never to have been used.
• The spurious inclusion of depend.cmi in the dependencies of ocamldep is a hangover from pre#1078 days

[dra27]

It was "wrong" before, but this should just be rm -f (no -r)

[dra27]

Typo: ocamlopt.opt.exe

[dra27]

I think it would be better to move these up with the bytecode version above (as is now the case for ocamlopt / ocamlopt.opt)

[dra27]

Stray line continuation

[dra27]

It reads strangely with 3 of these being prefixed with addprefix and ocamltoplevel not being. In this instance, might it be better just to have the list of 5 fully expanded (i.e. like the previous list, just without the extensions)

[dra27]

This gets corrected in a later commit, but this should be rm -f

[dra27]

I think it might be better to have $(TOOLS_TO_INSTALL) defined alongside the $(TOOLS) variable further up. It might also be clearer to put:

TOOLS_TO_INSTALL = ocamldep ocamlprof ocamlcp ..
TOOLS = $(TOOLS_TO_INSTALL) ocamlcmt dumpobj ...

?

[dra27]

I think this one doesn't need $(ROOTDIR). It looks like there a few tweaks needed to be able to say make lintapidiff - 5f2f478 seemed to be working for me (at least to get it to build!)

[shindere]

David Allsopp (2022/12/06 12:59 -0800):

@dra27 commented on this pull request. As ever, a thorough piece of work, and quite the nettle you've grasped again with this one, @shindere!

Many thanks for always being so supportive and, at the same time, so thorough in the reviews you are oding.

It's pleasing that a 353 line Makefile is removed with only 209 lines added elsewhere in its place :)

It is (perhaps the copyright header lines should not be counted as being removed though), many thanks for having spotted this.

I was concerned - incorrectly - that this might affect Cygwin64 when built in bootstrapped flexdll mode but this is working fine (I think that's due to a combination of your work on utils/Makefile and some of mine on $(MKEXE) et al).

That suggests (to me) that both works move us forward in a positive direction. Good news!

I also bumped all the magic numbers, and bootstrapping also seems to be working as expected.

Many thanks for having thought about doing this test!

I think you could squash the "Makefile.common: simplify the generic macros" into the previous commit, but it's fine if you want to keep the two separate versions.

No need to keep the two versions, you are right it's not worth keeping that bit of the history so I squashed, according to your suggestion.

Most of my comments are cosmetic or typos - the only major thing which needs fixing is the doubling of commands caused by ocamltoolsopt and ocamltoolsopt.opt both making the same recursive call to make in parallel in the opt.opt target. In passing: * In tools/ocamlmklib.ml, the syslib function is a leftover from ocamlbuild days and the chop_suffix functions appears never to have been used.

Oh, thanks for having checked the history for those!

* The spurious inclusion of depend.cmi in the dependencies of ocamldep is a hangover from pre#1078 days

And thanks for having checked this, too.

In [2]Makefile: > partialclean:: - rm -rf ocamlc$(EXE) + rm -rf ocamlc ocamlc.exe It was "wrong" before, but this should just be rm -f (no -r)

Well spotted (and fixed), thanks. I also noticed a line saying `rm $(ocamllex_PROGRAMS) $(ocamllex_PROGRAMS:=.exe)` (without a `-f`), so I added it, too, although its omission gets fixed by a later commit, as you have spotted in a comment that appears below.

In [3]Makefile: > partialclean:: - rm -f ocamlopt$(EXE) + rm -f ocamlopt ocamlopt.exe ocamlopt.opt ocmalopt.opt.exe Typo: ocamlopt.opt.exe

Fixed, thanks.

In [4]Makefile: > partialclean:: - rm -f ocamlc.opt$(EXE) - -# The native-code compiler compiled with itself - -ocamlopt.opt$(EXE): \ - compilerlibs/ocamlcommon.cmxa \ - compilerlibs/ocamloptcomp.cmxa \ - $(OPTSTART:.cmo=.cmx) - $(CAMLOPT_CMD) $(OC_COMMON_LDFLAGS) $(OC_NATIVE_LDFLAGS) -o $@ $^ - -partialclean:: - rm -f ocamlopt.opt$(EXE) + rm -f ocamlc.opt ocamlc.opt.exe I think it would be better to move these up with the bytecode version above (as is now the case for ocamlopt / ocamlopt.opt)

Done so.

In [5]Makefile: > @@ -1069,6 +1053,8 @@ partialclean:: # The lexer generator +ocamllex_LIBRARIES =\ Stray line continuation

Indeed, thanks. Fixed.

In [6]Makefile: > @@ -1335,27 +1318,26 @@ endif # The native toplevel -ocamlnat_dependencies := \ - compilerlibs/ocamlcommon.cmxa \ - compilerlibs/ocamloptcomp.cmxa \ - compilerlibs/ocamlbytecomp.cmxa \ - otherlibs/dynlink/dynlink.cmxa \ - compilerlibs/ocamltoplevel.cmxa \ - $(TOPLEVELSTART:.cmo=.cmx) +ocamlnat_LIBRARIES = \ + $(addprefix compilerlibs/ocaml,common optcomp bytecomp) \ It reads strangely with 3 of these being prefixed with addprefix and ocamltoplevel not being. In this instance, might it be better just to have the list of 5 fully expanded (i.e. like the previous list, just without the extensions)

Done so. I like it a bit less but I understand the concern so I did the change.

In [7]Makefile.common: > @@ -195,3 +195,34 @@ endif if_file_exists = ( test ! -f $(1) || $(2) && rm -f $(1) ) MERGEMANIFESTEXE = $(call if_file_exists, $(1).manifest, \ mt -nologo -outputresource:$(1) -manifest $(1).manifest) + +# Macros and rules to compile OCaml programs + +.SECONDEXPANSION: I think it's worth a comment here that .SECONDEXPANSION allows the _LIBRARIES and _MODULES definition to appear after the OCAML_PROGRAM call which uses them.

I have included (a variation of) the comment, thanks!

If we are being very pedantic, which we often are, then I think the .SECONDEXPANSION actually belongs two commits later in the history :)

I added it in the intermediate commit because it felt to me it was precisely its purpose to make this happen.

In [8]Makefile: > +$(foreach PROGRAM, $(C_PROGRAMS),\ + $(eval $(call PROGRAM_SYNONYM,$(PROGRAM)))) + +# OCaml programs that are compiled in both bytecode and native code + +OCAML_PROGRAMS = ocamlc ocamlopt lex/ocamllex + +$(foreach PROGRAM, $(OCAML_PROGRAMS),\ + $(eval $(call OCAML_PROGRAM,$(PROGRAM)))) + +# OCaml programs that are compiled only in bytecode +# Note: the bytecode toplevel, ocaml, is a bytecode program but at the +# moment it's a special one, because it needs to be expunged, so we +# can not declare it as we do for other bytecode-only programs. Typo: cannot rather than "can not"

Didn't know, but fixed, thanks.

In [9]Makefile: > @@ -444,11 +472,12 @@ ocaml.tmp: OC_BYTECODE_LDFLAGS += -I toplevel/byte -link all ocaml.tmp: $(ocaml_LIBRARIES:=.cma) $(ocaml_MODULES:=.cmo) $(LINK_BYTECODE_PROGRAM) -o $@ $^ +$(eval $(call PROGRAM_SYNONYM,ocaml)) Is this worth putting with OCAML_BYTECODE_PROGRAMS above? It goes well with the comment about why the ocaml command is a special case?

As discussed off-line, I am suggesting to leave the re-ordering of the root Makefile for later, since it feels a task worth doing independently. Is that okay?

In [10]Makefile.build_config.in: > @@ -108,10 +108,17 @@ DEFAULT_BUILD_TARGET = @default_build_target@ OC_COMMON_CFLAGS = -g -strict-sequence -principal -absname \ -w +a-4-9-40-41-42-44-45-48 -warn-error +a -bin-annot \ -strict-formats -OC_COMMON_LDFLAGS = -g +OC_COMMON_LDFLAGS = -g $(INCLUDES) + +OC_BYTECODE_LDFLAGS = Should these be in Makefile.common if they're not actually being assigned values from configure?

As discussed offline, it is hoped that, one day, these flags will be controlled by configure, e.g. to make it possible to control the inclusion of debugging information from configure. I could add that, if we reach the stage where there will be only one Makefile, then `Makefile.common` will have no more reason to exist and may then disappear.

In [11]tools/ocamltex.ml: > @@ -207,15 +207,16 @@ module Toplevel = struct Location.input_name := fname; Location.input_lexbuf := Some lex; try + let module P = Parsetree in I think we could cope with the two instances of Parsetree below and just not have this line? :)

Yes, you are right that it's better that way. After all there are no short names for the other modules either so there is no good reason to have one for this one.

In [12]tools/ocamltex.mli: > @@ -0,0 +1,2 @@ +(* ocamltex.ml does not need to export any definition so here is + an empty interface *) Please could this one have a copyright header, too?

Sure, done so, sorry I forgot.

In [13]tools/lintapidiff.mli: > @@ -0,0 +1,2 @@ +(* lintapidiff.ml does not need to export any definition so here is + an empty interface *) Copyright header

Done

In [14]tools/make_opcodes.mli: > @@ -0,0 +1 @@ +(* make_opcodes.mll exports no definition so here is an empty interface *) Copyright header

Done

In [15]Makefile: > middle_end/closure middle_end/flambda middle_end/flambda/base_types \ - asmcomp driver toplevel -INCLUDES = $(addprefix -I ,$(DIRS)) + asmcomp driver toplevel tools +INCLUDES = $(addprefix -I ,$(VPATH)) The need for -I space could be made explicit with -I$(SPACE)? (it obviously wasn't before)

Frankly, I am unsure. To me the rule is readable as is so unless you insist I'd tend to leave it alone, at least for the time being.

In [16]Makefile: > @@ -416,37 +407,42 @@ clean:: # The clean target clean:: partialclean rm -f configure~ $(PROGRAMS) $(PROGRAMS:=.exe) + rm $(ocamllex_PROGRAMS) $(ocamllex_PROGRAMS:=.exe) This gets corrected in a later commit, but this should be rm -f

Yes, you are totally correct. I still fixed the commit, as discussed earlier.

In [17]Makefile: > +TOOLS_TO_INSTALL = \ + ocamldep ocamlprof ocamlcp ocamloptp ocamlmktop ocamlmklib ocamlobjinfo I think it might be better to have $(TOOLS_TO_INSTALL) defined alongside the $(TOOLS) variable further up. It might also be clearer to put: TOOLS_TO_INSTALL = ocamldep ocamlprof ocamlcp .. TOOLS = $(TOOLS_TO_INSTALL) ocamlcmt dumpobj ...

Done. I also introduced a `TOOLS_PROGRAMS` that has similar content to `TOOLS` but with the `tools/` prefix added. Note that I started by calling that variable `TOOL_PROGRAMS` which seemed more grammatical but then I went for `TOOLS_PROGRAMS` to refer to the fact that it talks about programs that are in the `tools` directory. And, while I was at it, I also introduced `TOOLS_MODULES` to avoid having to repeat `tools/profiling` in the definitions of `TOOLS_BYTECODE_TARGETS` and `TOOLS_NATIVE_TARGETS`. Thisalso makes it eaiy to add other tool modules, should we need to.

In [18]Makefile: > .PHONY: ocamltoolsopt ocamltoolsopt: ocamlopt - $(MAKE) -C tools opt + $(MAKE) tools-allopt This isn't quite right - before this target compiled just tools/profiling.cmx but now it does a full build and because it's in a recursive make call, things are being built twice. It's sufficient to fix the build just by removing the call to ocamltoolsopt in the opt.opt target, but this technically means that make world && make opt now compiles the .opt versions of the tools, which it's not supposed to.

I am unsure how to go about this, then.

In [19]Makefile: > else checkstack: @ endif # Lint @SInCE and @deprecated annotations +lintapidiff_LIBRARIES = \ + $(addprefix compilerlibs/ocaml,common bytecomp middleend) \ + otherlibs/str/str +lintapidiff_MODULES = tools/lintapidiff + +tools/lintapidiff.opt$(EXE): VPATH += $(ROOTDIR)/otherlibs/str I think this one doesn't need $(ROOTDIR).

You are right indeed, thanks. Fixed.

It looks like there a few tweaks needed to be able to say make lintapidiff

There was a mistake in the definition of `OCAML_NATIVE_PROGRAMS`: it contained `tools/lintapidiff` but should have contained `tools/lintapidiff.opt` for the macro to do the right things. That happened because the way to deal with the `.opt` suffix has change during this PR has been worked on. It's now fixeed and seems to work, at least as far as building `tools/lintapidiff.opt` is concerned.

In [21]Makefile: > + +cmpbyt_LIBRARIES = $(addprefix compilerlibs/ocaml,common bytecomp) +cmpbyt_MODULES = tools/cmpbyt + +# Scan latex files, and run ocaml code examples + +ocamltex_LIBRARIES = \ + $(addprefix compilerlibs/ocaml,common bytecomp toplevel) \ + $(addprefix otherlibs/,str/str unix/unix) +ocamltex_MODULES = tools/ocamltex + +# ocamltex uses str.cma and unix.cma and so must be compiled with +# $(ROOTDIR)/ocamlc rather than with $(ROOTDIR)/boot/ocamlc since the boot +# compiler does not necessarily have the correct shared library +# configuration. +$(ocamltex): CAMLC = $(OCAMLRUN) $(ROOTDIR)/ocamlc$(EXE) $(STDLIBFLAGS) I think it would be worth adding a comment to note that this also applies to the compilation of ocamltex.ml and ocamltex.mli (because they are dependencies) as that's both important and also a subtle part of how GNU make works

I did add the comment. It's jsut that the definitions apply to all theprerequisites so even to the libraries I think, so I am wondering whether this is quite right. What do you think?

[dra27]

All good, thanks @shindere! Two specific replies:

for the issue with the opt and opt.opt targets:

I am unsure how to go about this, then.

I'll pull the revised branch and have a look (apropos our offline discussion)

for the overriding of CAMLC in ocamltex's compilation:

I did add the comment. It's jsut that the definitions apply to all theprerequisites so even to the libraries I think, so I am wondering whether this is quite right. What do you think?

Hmm, it's kind of a dirty "secret" of the build at the moment that compilerlibs is compiled using $(ROOTDIR)/boot/ocamlc but has to be linkable by programs compiled with $(ROOTDIR)/ocamlc. Indeed, at the moment, if you do a build, then, say, touch parsing/location.ml tools/ocamltex.ml ; make tools/ocamltex then ocamlcommon.cma gets incorrectly rebuilt using $(ROOTDIR)/ocamlc instead of $(ROOTDIR)/boot/ocamlc. That can be changed by having an explicit compilerlibs/ocamlcommon.cma: CAMLC = ... in compilerlibs/Makefile.compilerlibs - perhaps that a good idea, therefore, to make the compiler used for the libraries explicit?

I applied this inelegant copy-and-paste diff:

--- a/compilerlibs/Makefile.compilerlibs
+++ b/compilerlibs/Makefile.compilerlibs
@@ -416,6 +416,7 @@ utils/config_%.mli: utils/config.mli

 beforedepend:: utils/config_main.mli utils/config_boot.mli

+compilerlibs/ocamlcommon.cma: CAMLC = $(BOOT_OCAMLC) $(BOOT_STDLIBFLAGS) -use-prims runtime/primitives
 compilerlibs/ocamlcommon.cma: $(COMMON_CMI) $(ALL_CONFIG_CMO) $(COMMON)
        $(CAMLC) -a -linkall -o $@ $(COMMON)
 partialclean::

and then did the touch parsing/location.ml ... from above and it looks like location.cmo and ocamlcommon.cma are correctly produced using $(ROOTDIR)/boot/ocamlc and then ocamltex.cmo and ocamltex using $(ROOTDIR)/ocamlc:

dra@thor:~/ocaml-7$ touch parsing/location.ml tools/ocamltex.ml
dra@thor:~/ocaml-7$ make tools/ocamltex
./boot/ocamlrun ./boot/ocamlc -nostdlib ... -c parsing/location.ml
./boot/ocamlrun ./boot/ocamlc -nostdlib ... -o compilerlibs/ocamlcommon.cma ...
./boot/ocamlrun ./ocamlc -nostdlib ... -c tools/ocamltex.ml
./boot/ocamlrun ./ocamlc -nostdlib ... -o tools/ocamltex ...

It may also be possible to use a pattern rule to apply that a bit more elegantly (I think that might even be stable with the newer prefix-length, rather than order-in-makefile, rules for applying pattern rules?)

[Octachron]

Wrong copyright?

[dra27]

Typo: extra S at the end of TOOLS_PROGRAMS

[shindere]

Florian Angeletti (2022/12/09 02:06 -0800):

@Octachron commented on this pull request. __________________________________________________________________ In [1]tools/objinfo.mli: > @@ -0,0 +1,17 @@ +(**************************************************************************) +(* *) +(* OCaml *) +(* *) +(* Damien Doligez and Francois Rouaix, INRIA Rocquencourt *) +(* Ported to Caml Special Light by John Malecki *) Wrong copyright?

Yes, absolutely. Sorry about that and many thanks for having spotted it. The copyright was also wrong in a few other files, that should now be in order.

[shindere]

I also added you to the Changes entry, @Octachron.

[shindere]

David Allsopp (2022/12/09 02:12 -0800):

@dra27 commented on this pull request. __________________________________________________________________ In [1]Makefile: > git ls-files -- 'otherlibs/*/*.mli' 'stdlib/*.mli' |\ grep -Ev internal\|obj\|stdLabels\|moreLabels |\ tools/lintapidiff.opt $(VERSIONS) # Tools +TOOLS_BYTECODE_TARGETS = $(TOOLS_PROGRAMS) $(TOOLS_MODULES:=.cmo) + +TOOLS_NATIVE_TARGETS = $(TOOLS_PROGRAMSS:=.opt) $(TOOLS_MODULES:=.cmx) Typo: extra S at the end of TOOLS_PROGRAMS

Oops, thanks. Fixed.

[shindere]

David Allsopp (2022/12/09 01:46 -0800):

All good, thanks @shindere! Two specific replies: for the issue with the `opt` and `opt.opt` targets: > I am unsure how to go about this, then. I'll pull the revised branch and have a look (apropos our offline discussion)

Many thanks. For the record, @dra27 has proposed dra27@d6867e6, which I thought I could improve by doing some computations during the configure stage but I didn't like the result so @dra27 kindly agreed that his commit be cherry-picked as part of this pull request.

for the overriding of `CAMLC` in `ocamltex`'s compilation: > I did add the comment. It's jsut that the definitions apply to all theprerequisites so even to the libraries I think, so I am wondering whether this is quite right. What do you think? Hmm, it's kind of a dirty "secret" of the build at the moment that compilerlibs is compiled using `$(ROOTDIR)/boot/ocamlc` but has to be linkable by programs compiled with `$(ROOTDIR)/ocamlc`.

My understanding is that will no longer be true once the work on the build system has been completed. At least that this will not have to be true. Do you also understand things that way, @dra27?

Indeed, at the moment, if you do a build, then, say, `touch parsing/location.ml tools/ocamltex.ml ; make tools/ocamltex` then `ocamlcommon.cma` gets incorrectly rebuilt using `$(ROOTDIR)/ocamlc` instead of `$(ROOTDIR)/boot/ocamlc`. That can be changed by having an explicit `compilerlibs/ocamlcommon.cma: CAMLC = ...` in `compilerlibs/Makefile.compilerlibs` - perhaps that a good idea, therefore, to make the compiler used for the libraries explicit?

But then,

I applied this inelegant copy-and-paste diff: ```diff --- a/compilerlibs/Makefile.compilerlibs +++ b/compilerlibs/Makefile.compilerlibs @@ -416,6 +416,7 @@ utils/config_%.mli: utils/config.mli beforedepend:: utils/config_main.mli utils/config_boot.mli +compilerlibs/ocamlcommon.cma: CAMLC = $(BOOT_OCAMLC) $(BOOT_STDLIBFLAGS) -use-prims runtime/primitives compilerlibs/ocamlcommon.cma: $(COMMON_CMI) $(ALL_CONFIG_CMO) $(COMMON) $(CAMLC) -a -linkall -o $@ $(COMMON) partialclean:: ``` and then did the `touch parsing/location.ml ...` from above and it looks like `location.cmo` and `ocamlcommon.cma` are correctly produced using `$(ROOTDIR)/boot/ocamlc` and then `ocamltex.cmo` and `ocamltex` using `$(ROOTDIR)/ocamlc`: ``` ***@***.***:~/ocaml-7$ touch parsing/location.ml tools/ocamltex.ml ***@***.***:~/ocaml-7$ make tools/ocamltex ./boot/ocamlrun ./boot/ocamlc -nostdlib ... -c parsing/location.ml ./boot/ocamlrun ./boot/ocamlc -nostdlib ... -o compilerlibs/ocamlcommon.cma ... ./boot/ocamlrun ./ocamlc -nostdlib ... -c tools/ocamltex.ml ./boot/ocamlrun ./ocamlc -nostdlib ... -o tools/ocamltex ... ``` It may also be possible to use a pattern rule to apply that a bit more elegantly (I think that might even be stable with the newer prefix-length, rather than order-in-makefile, rules for applying pattern rules?)

Do you mean to apply this to all the bytecode libraries built in compilerlibs? So something like ``` compilerlibs/%.cma: CAMLC = $(BOOT_OCAMLC) $(BOOT_STDLIBFLAGS) -use-prims runtime/primitives ``` ? By the way: except if we want to add the presently discussed change, the PR should now be ready for a last review and, maybe, to be merged.

[dra27]

tools/ocamloptp needs adding to this list (as it's used for the clean target) - this is the cause of the CI failure at the moment

[dra27]

My understanding is that will no longer be true once the work on the
build system has been completed. At least that this will not have to be
true. Do you also understand things that way, @dra27?

Yes, that's right - I was anticipating that at some point we'll be able to tell trivially whether a separate set of compiler-libs is needed, or not.

Do you mean to apply this to all the bytecode libraries built in
compilerlibs? So something like

compilerlibs/%.cma: CAMLC = $(BOOT_OCAMLC) $(BOOT_STDLIBFLAGS) -use-prims runtime/primitives

Yes (apart from the clean check which is failing in CI) - I just don't know (without trying) how the pattern rules will interact.

By the way: except if we want to add the presently discussed change, the
PR should now be ready for a last review and, maybe, to be merged.

Yes, I agree! I think it might be worth trying to fix this issue here, though - it does at least sit right at the end of the commit series.

[shindere]

David Allsopp (2022/12/12 04:10 -0800):

`tools/ocamloptp` needs adding to this list (as it's used for the `clean` target) - this is the cause of the CI failure at the moment

Indeed, sorry! Fixed.

[shindere]

David Allsopp (2022/12/12 04:13 -0800):

> Do you mean to apply this to all the bytecode libraries built in compilerlibs? So something like ``` compilerlibs/%.cma: CAMLC = $(BOOT_OCAMLC) $(BOOT_STDLIBFLAGS) -use-prims runtime/primitives ``` Yes (apart from the `clean` check which is failing in CI) - I just don't know (without trying) how the pattern rules will interact.

Okay so I pushed one commit to fix `make clean`. As discussed, `tools/ocamloptp` etc. are removed unconditionnally because during `make clean` the configuration is not included, even if it exists.

> By the way: except if we want to add the presently discussed change, the PR should now be ready for a last review and, maybe, to be merged. Yes, I agree! I think it might be worth trying to fix this issue here, though - it does at least sit right at the end of the commit series.

I have pushed another commit that defines the right compiler to use for both `ocamlcommon.cma` and `ocamlbytecomp.cma`. CI is happy now. The AppVeyor failure is unrelated (one memory model test failed). So hopefully this should finally be ready to go. IN any case many thanks, @dra27 for allyour support.

[dra27]

This is making a journey through precheck#807 and assuming that doesn't show up anything, this is good to go with CI!

[dra27]

The mingw32 failure in Jenkins is a sporadic memory model test failure - this is ready to go. A.OUT¹!

Footnotes

1. see last section of https://www.bell-labs.com/usr/dmr/www/odd.html ↩