On Fri, Sep 30, 2022, at 9:24 AM, Chet Ramey wrote: > On 9/30/22 1:45 AM, Xi Ruoyao wrote: >> >> Hmm, it makes sense but I can't see how this affects autoconf test suite >> (I don't really understand M4). > > It might also be the result of a different change that optimizes out some > unnecessary forks; there are a couple of those. I'm not sure why autoconf > is keeping track of SHLVL anyway, unless it's recording state. Autoconf's test suite is testing for Autoconf macros that modify shell variables that are "outside the Autoconf namespace" (see https://git.savannah.gnu.org/cgit/autoconf.git/tree/tests/local.at#n320 for further explanation). We already have a long list of built-in shell variables that may change without our code having done anything, and so should be ignored by this check. The only reason SHLVL isn't already in that list is because, hitherto, it has consistently had the same value before and after each of our top-level macros. I'm a little puzzled why that's no longer true, but clearly it's not a bug _in Autoconf_ and shouldn't be flagged by our test suite. Xi's original patch adds SHLVL to that list and I'm going to go ahead and apply it. zw
