added -threads and -vmthreads option information to the ppx context

[camlspotter]

Background

PPX preprocessors which search existing modules, such as ppx_overload, ppx_implicits and ppx_import, use Config.load_path as the search path. Config.load_path is calculated by Compmisc.init_path using Clflags.include_dirs, Clflags.use_threads and Clflags.use_vmthreads.

Currently only Clflags.include_dirs is passed from the host compiler command to PPX preprocessors: this makes them running with Config.load_path different from the one of the host compiler when -thread or -vmthread is given to the host compiler. As a result, PPX preprocessors cannot find thread related modules only by calling Compmisc.init.

The patch

This patch adds Clflags.use_threads and Clflags.use_vmthreads to the PPX context so that PPX preprocessors can configure itself with the same load path as the host compiler.

[alainfrisch]

This, and the copyright notice below, do not seem right.

[camlspotter]

Ouch, a stuipid copy and paste.

I am going to remove it, or do you need a copy right notice of my name ?

[mshinwell]

My understanding is:

• If this Makefile was pilfered from somewhere else, you should preserve the original author attribution, and add your name and copyright if you made non-trivial modifications.
• If this Makefile is your own work, then it should have your name and copyright (and no-one else's to begin with).

In both cases they should have the LGPL paragraph.

I'm suspecting something along these lines didn't happen correctly at a previous time, since this Makefile has an INRIA copyright yet a non-INRIA personal affiliation.

[alainfrisch]

There is no copyright header in the testsuite tests themselves. @mshinwell Do you know why we keep them in the testsuite Makefiles?

[mshinwell]

I do not.

[alainfrisch]

@damiendoligez @xavierleroy What is the official recommendation w.r.t. copyright headers in the testsuite? Currently they are kept in Makefiles but not in tests themselves. Should we get rid of them in Makefiles as well (or at least not require them for new Makefiles)?

[damiendoligez]

No copyright headers in the test suite except for test infrastructure scripts and Makefiles. In general, the test directories themselves should not have any headers, even in Makefiles. I just haven't gotten around to remove them yet (and I won't need to thanks to ocamltest).

[alainfrisch]

Ok, so @camlspotter : can you just remove the header?

[alainfrisch]

This looks good to me, the PPX could make use of the knowledge about whether the code is compiled with -thread/-vmthread for other purposes.

What I don't immediately see is why this is required for load_path to be correct, since load_path itself (not just include_dirs) is also passed to the ppx in the context. But perhaps something within the ppx recomputes load_path from include_dirs and other flags?

[camlspotter]

Ah now I understand. Config.load_path is transferred via PPX contexts, as you wrote, but Compmisc.init_path resets it. If PPX performs some typing using Compile.interface or Compile.implementation, it calls Compmisc.init_path and the issue happens.

Without this patch, I can avoid the issue by saving the value of Config.load_path before PPX calls Compmisc.init_path then pushing the value back after Compmisc.init_path. But it requires dirty workaround coding and I would like to avoid it if possible.

[alainfrisch]

Ok, fair enough. I guess that one could arrange not to destroy load_path when running in the context of a PPX, but I'm not sure this would be more satisfactory than the proposed solution.

If nobody else objects to it, I'm tempted to merge (after sorting out what needs to be done with the Makefile copyright notice).

[shindere]

Alain Frisch (2017/09/15 12:30 +0000):

alainfrisch commented on this pull request. > @@ -0,0 +1,41 @@ +#************************************************************************** +#* * +#* OCaml * +#* * +#* Peter Zotov * There is no copyright header in the testsuite tests themselves. @mshinwell Do you know why we keep them in the testsuite Makefile?

I'd say it's not a big deal. These makefiles will vanish slowly and their headers with them so let's let them die "de leur belle mort".

[mshinwell]

@alainfrisch I think you can just push to camlspotter's branch to remove the header and merge.

[mshinwell]

I did the merge with trunk myself. I also removed the offending copyright header and removed some trailing whitespace. I will merge this to trunk now.

[camlspotter]

Thanks!