Re: accessing ENVIRON causes SIGFPE when first var is numeric

[arnold]

"David Arroyo" <david@aqwari.net> wrote:

> Thanks for your response!

You're welcome.

> My distribution does not ship HEAD, it is at version 5.1.0. I reproduced
> at HEAD to demonstrate that the issue is not resolved upstream.

Ah, I see.  Thanks!

> I also did not expect my distribution to set numeric environment
> variables, but I think it's a result you can get accidentally from running
> your shell with `set -a`. If it's not posix compliant I will report it.

Great. Andy sent a better patch that I will use.

Thanks again for the report.

Arnold

> Thanks,
> David
>
> On Tue, Jun 14, 2022, at 20:35, arnold@skeeve.com wrote:
> > Hi.
> >
> > Thanks for the report.  Your fix is what I would have done; I will
> > get it into the code base.
> >
> > As a general remark, using HEAD from masster, that is, an unreleased
> > version, isn't necessarily recommended.  I hope that's not what
> > your distribution is shipping but just what you happened to build.
> >
> > Also, the use of a purely numeric value as an environment variable
> > name is at the least against standard practice; it might even
> > violate POSIX, but I gather this is what your distribution does
> > for you.
> >
> > In any case, much thanks!
> >
> > Arnold
> >
> > "David Arroyo" <david@aqwari.net> wrote:
> >
> >> Configuration Information [Automatically generated, do not change]:
> >> Machine: x86_64
> >> OS: linux-gnu
> >> Compiler: gcc
> >> Compilation CFLAGS: -g -O2 -DNDEBUG
> >> uname output: Linux dorcas 5.15.43 #1 SMP 1 x86_64 GNU/Linux
> >> Machine Type: x86_64-pc-linux-gnu
> >>
> >> Gawk Version: 5.1.60 (HEAD)
> >>
> >> Attestation:
> >>         I have read 
> >> https://www.gnu.org/software/gawk/manual/html_node/Bugs.html.
> >>         Yes
> >>
> >> Description:
> >>
> >> Since commit 7d19cbd5, if the first environment variable is a
> >> number, gawk will treat ENVIRON as a cint_array instead of a
> >> str_array. You can reproduce this with the following command
> >> line:
> >>
> >>    env -i \
> >>      $(which env) 0=foo x=bar \
> >>      $(which gawk) 'BEGIN { ENVIRON["x"] }' /dev/null
> >>
> >> Will yield the error:
> >>
> >>    gawk: cmd. line:1: fatal: floating point exception
> >>
> >>  at str_array.c:607 
> >>
> >> And I can see from a debugger that after the first iteration
> >> of load_environ, ENVIRON_node->array_funcs becomes
> >> cint_array_func. This issue does not occure when using
> >> the --posix flag.
> >>
> >> I ran into this error on my personal computer running GNU
> >> Guix, where the first variable in my environment is
> >>
> >>    0=/home/me/bin/startw
> >>
> >> I am not sure if this is common, as I do not see such an
> >> env var on a debian 10 system I had handy, and I have not
> >> tracked down what exactly sets that variable. But it seems
> >> undesirable for variations in user environment variables
> >> to trigger a crash.
> >>
> >> Fix:
> >>
> >> The patch below hardcodes ENVIRON_node to use the
> >> str_array_func family of functions. I don't know if that is
> >> the best way to fix the issue, but it helps to illustrate what's
> >> happening. I could not reproduce this issue using an 
> >> arbitrary array.
> >>
> >> ---
> >>  main.c | 1 +
> >>  1 file changed, 1 insertion(+)
> >>
> >> diff --git a/main.c b/main.c
> >> index 2cdd1e2d..3f48ddfb 100644
> >> --- a/main.c
> >> +++ b/main.c
> >> @@ -922,6 +922,7 @@ load_environ()
> >>    been_here = true;
> >>
> >>    ENVIRON_node = install_symbol(estrdup("ENVIRON", 7), Node_var_array);
> >> +  ENVIRON_node->array_funcs = &str_array_func;
> >>    for (i = 0; environ[i] != NULL; i++) {
> >>            static char nullstr[] = "";
> >>
> >> --
> >>