Re: [PATCH] config.sub: go back to backtick substitutions

[Jacob Bachmeyer]

Ozkan Sezer wrote:
> On 7/3/21, Jacob Bachmeyer <jcb62281@gmail.com> wrote:
>   
> > Ozkan Sezer wrote:
> >     
> > > The attached patch replaces a handful of POSIX $(xx) with classic `xx`
> > > in config.sub. This is in line with the recent config.guess change i.e.
> > > commit d70c4fa934de:
> > > http://git.savannah.gnu.org/gitweb/?p=config.git;a=commit;h=d70c4fa934de164178054c3a60aaa0024ed07c91
> > >
> > > --
> > > O.S.
> > >
> > >       
> > While doing this, also normalizing the quoting in the `echo FOO | sed
> > ...` would be nice.  Some cases quote the argument to echo and some do
> > not.  At runtime it probably does not matter because the substituted
> > values will never contain whitespace, but quoting them all would make
> > shellcheck more useful.
> >
> > Also see patch 5 in the recent "v2" series for the marker line to
> > prevent shellcheck from complaining about backticks.  It must be placed
> > before the first executable line (the timestamp variable assignment) for
> > shellcheck to apply it to the entire script.
> >
> >
> > -- Jacob
> >     
>
> Done, as best as I can: Attached patches.
>
> $ make check-sub
> cd testsuite && bash config-sub.sh
> PASS: config.sub checks (864 tests)
> PASS: config.sub idempotency checks (801 tests)
> PASS: config.sub canonicalise each config.guess testcase (136 tests)
>
> shellcheck barfs with SC1117 in config.guess, but no errors with
> config.sub.
>   

I get no errors with either of them when pasting them at shellcheck.net 
as of commit 45e181800a6a27268a9c5d79dcc60492fef9a9a0.  Looking up 
SC1117 shows that you are using a very old shellcheck -- that error was 
retired due to excessive pedantry.  (Which is also how I view the 
complaints about backticks and expr...)

> Off-topic: check-sub fails on my old setup with CentOS-6.10:
> config-sub.sh: line 12: shopt: lastpipe: invalid shell option name
>   

That is part of the reason that I did not already write a patch for 
config.sub:  when I last checked, that test script also failed on 
fencepost!  (A server available to GNU maintainers (I now maintain 
DejaGnu, which uses config.guess) as a reference GNU platform among 
other uses.  Code that does not work there is probably "a bit" bleeding 
edge for general use.)

If there is interest, I could puzzle out what that script actually does 
and look into writing a patch to make it more compatible with 
not-quite-absolute-latest bash, but you seem to be taking care of the 
major change I would consider useful to config.sub.  :-)  Thanks.


-- Jacob