Re: nodejs fails on make install -current i386

[Greg Troxel <gdt%lexort.com@localhost>]

Phil Rulon <pjr%zephyrite.net@localhost> writes:

> Greg Troxel writes:
>> FWIW, on netbsd-8, amd64, with a lot of other things installed, and
>> up-to-date lang/nodejs, it built and packaged ok for me (with
>> PKG_DEVELOPER=yes).
>
> I also saw no issues with my amd64 machines running -8.  The problem seems to
> be isolated to i386.  There is a bit of assembler code with C fallbacks that
> may not be getting switched in properly.  This may not be the source of the
> problem.

I was able to reproduce a failure to package on 

> What's notable is that the make phase runs fine.  It's only when make install
> is run, that missing files are discovered.  I've looked at the code a bit,
> this package is one of the relatively hairy ones, so far no joy.  I did try it
> with PKG_DEVELOPER=yes.  I'm not much of an autotools guy, so it's slow going.

The make phase emits an error.  So there are two things wrong: using the
presumably non-POSIX "-not", and failing to check exit status of find.

A quick grep over the tree yields:

  ./work/node-v10.14.2/deps/v8/tools/android-sync.sh:  for FILE in $(find "$HOST_V8/$DIR" -not -path "*.svn*" -type f); do
  ./work/node-v10.14.2/deps/v8/tools/check-inline-includes.sh:  headers=$(find "$v8_root/$directory" -name '*.h' -not -name '*-inl.h')

So probably that's the issue.

You mith try changing to \! which is I think the standard form.

> I'm a little worried that folks are giving up on the i386, I hope not.  I still
> have an old ThinkPad that I'd like to run a browser on.

Well, there is some of that going on.  Some upstreams are not paying
attention, and some things are getting harder on i386 because of
upstream choices.  In pkgsrc, people more or less try to keep it working
everywhere, but there's only so much effort and sometimes it doesn't
work.