Re: segfault of bison with some parser input

[Akim Demaille]

Le 9 mars 2011 à 21:13, Akim Demaille a écrit :

> For the records, these are the patches as installed on 2.5 (including the 
> change of address Tys asked me).

Now for master.

The first patch is identical, but the second one (the real one) is slightly 
different because of changes in parse-gram.y.

0002-named-references-fix-double-free.txt
Description: Text document

It took me a while to test this patch, because it seems that bootstrapping a 
checkout of master on top of branch-2.5 does not work properly.  I had to 
remove lib/gnulib.mk in order to have bootstrap extract it and run our 
conversion program.  Otherwise, apparently, bootstrap thinks lib/gnulib.mk is 
nice looking as it is, and decides not to install it, and therefore, not to 
apply our gnulib_mk_hook.

      if test $file = Makefile.am && test "X$gnulib_mk" != XMakefile.am; then
        copied=$copied${sep}$gnulib_mk; sep=$nl
        remove_intl='/^[^#].*\/intl/s/^/#/;'"s!$bt_regex/!!g"
        sed "$remove_intl" $1/$dir/$file |
        cmp - $dir/$gnulib_mk > /dev/null || {
          echo "$0: Copying $1/$dir/$file to $dir/$gnulib_mk ..." &&
          rm -f $dir/$gnulib_mk &&
          sed "$remove_intl" $1/$dir/$file >$dir/$gnulib_mk &&
          gnulib_mk_hook $dir/$gnulib_mk
        }

cmp is not comparing the same thing: one is transformed, the other one is not.  
In 2.5 it does not need to be transformed, so it is not, and when moving to 
master, where the transformation is needed, it thinks all is done.

We should probably remove gnulib.mk at the beginning of bootstrap.

(and bummer, I just saw wrote rhs where I meant lhs, in the log... bah)