Re: Fix subst SUBST_NOOP_OK?=yes fallout

[Jonathan Perkin <jperkin%joyent.com@localhost>]

* On 2020-06-23 at 19:32 BST, Roland Illig wrote:

> On 23.06.2020 20:01, Jonathan Perkin wrote:
> > * On 2020-06-23 at 18:28 BST, Roland Illig wrote:
> > 
> > > Can you run the following lines before and after the fix? Just to make
> > > sure that the subst.mk test suite would have caught this pdksh bug.
> > > 
> > > cd regress/infra-unittests
> > > sh subst.sh
> > 
> > Before the fix everything fails (44 "assertion failed in ...").
> > 
> > After the fix I still see some failures:
> > 
> >    $ pdksh subst.sh
> >    assertion failed in "brackets in filename patterns": file "*" has unexpected content:
> >    --- /tmp/infra-unittests-47062/expected
> >    +++ *
> >    @@ -1 +1 @@
> >    -after
> >    +before
> > 
> >    assertion failed in "brackets in filename patterns": file "[*]" has unexpected content:
> >    --- /tmp/infra-unittests-47062/expected
> >    +++ [*]
> >    @@ -1 +1 @@
> >    -before
> >    +after
> 
> Oh, that's funny. This means that your shell does not expand the square
> brackets in this code:
> 
> pattern='[*]'
> for filename in $pattern; do
>   :
> done
> 
> All of NetBSD-sh, shells/bash, shells/dash, shells/mksh, shells/pdksh,
> shells/oksh do that, though.
> 
> Assuming that shells/pdksh is your "primary shell", what is the shell
> used by bmake? It would be strange if pdksh behaved differently between
> NetBSD and macOS.

I recently committed these:

  https://github.com/NetBSD/pkgsrc/commit/94a7dba7283c
  https://github.com/NetBSD/pkgsrc/commit/740ff71b71f5

which means it is now also shells/pdksh, but prior to those commits it
was /bin/sh (aka bash), and would be for most OS, which probably
explained why my first pdksh bulk build (performed before these were
committed) looked ok.

> >    assertion failed in "unreadable file": file "/tmp/infra-unittests-47062/output" has unexpected content:
> >    --- /tmp/infra-unittests-47062/expected
> >    +++ /tmp/infra-unittests-47062/output
> >    @@ -1,7 +1 @@
> >     => Substituting "id" in unreadable-file
> >    -sh: cannot open unreadable-file: permission denied
> >    -sh: cannot open unreadable-file: permission denied
> >    -*** Error code 1
> >    -
> >    -Stop.
> >    -bmake: stopped in /tmp/infra-unittests-47062/work
> 
> Ah, I didn't think that someone would run the tests as root. Obviously
> root doesn't care about the file mode.

Builds within pbulk are built by the pbulk user, but for general
development work and testing I just use root as its far simpler (and
obviously still fully sandboxed).

-- 
Jonathan Perkin  -  Joyent, Inc.  -  www.joyent.com