Control: reassign 1112461 eperl Control: severity -1 important Control: forcemerge -1 1112461 Control: tags -1 + patch
[CC += Niko in case this is actually a perl bug and I'm not holding it wrong,
please cf. section starting "Steps to repro"]
I can reproduce this.
However: this does /not/ appear to be an eperl thing;
or at least, not exclusively.
I'm running the same configuration in a bookworm chroot
and a trixie chroot, and I see:
bookworm bookworm bin:eperl full output
bookworm trixie bin:eperl (rebuilt) full output
bookworm git ossp-eperl 2.2.15 full output
trixie trixie bin:eperl no output
trixie bookworm bin:eperl (rebuilt) full output
trixie git ossp-eperl 2.2.15 no output
(where "no output" is one line, like it is for you, but that line is
GATEWAY_INTERFACE = CGI/1.1
so just one, but a different one
So this looks like some interaction between eperl
/and/ the rest of the system
(apache2? CGI.pm? perl 2.38 -> 2.40?).
(apache2? probably not, a CGI of
#!/bin/sh
echo content-type: text/plain
echo
exec env
is the same (sans like PATH and SERVER_SOFTWARE)
between bookworm and trixie.)
I get full output on trixie + trixie bin:perl
when I run eperl -w instead of eperl
(through a shell "exec eperl ..." wrapper).
Likewise for -x.
...
I've actually isolated the repro to
env -i GATEWAY_INTERFACE='CGI/1.1' PATH_TRANSLATED='/srv/a.phtml' ./eperl
where adding -x, -w, /or/ -mc makes it work.
-mc is notable because this is /already/ CGI mode.
I can't otherwise ascertain any meaningful difference.
...actually even passing an empty string as the first argument works.
what the hell.
There appears to be no effect to this that I can see. Except that it
makes it work. Hell, even
if(!argv[optind])
argv[optind] = "";
after the getopt loop fixes it(!?):
-- >8 --
diff --git a/eperl_main.c b/eperl_main.c
index 13417db..237b093 100644
--- a/eperl_main.c
+++ b/eperl_main.c
@@ -347,6 +347,9 @@ int main(int argc, char **argv)
return EX_USAGE;
}
+if(!argv[optind])
+argv[optind]="";
+
/*
* determine source filename and runtime mode
*/
-- >8 --
But only with perl 5.40. I can't reproduce this on perl 5.36.
Steps to repro:
1. git clone https://git.sr.ht/~nabijaczleweli/ossp-eperl
(I'm testing @ 97313d69c99b72863b9dd23c442c4fa881b483b3
which is equivalent to ePerl_2_2_15 and src:eperl 2.2.15-1)
2. autoreconf -fi && ./configure && make eperl
3. echo '<? foreach my $var (keys %ENV) {print "$var = $ENV{$var}\n"; } !>' >
b.phtml
4. env -i GATEWAY_INTERFACE=CGI/1.1 PATH_TRANSLATED="$PWD/b.phtml" ./eperl
5. env -i GATEWAY_INTERFACE=CGI/1.1 PATH_TRANSLATED="$PWD/b.phtml" ./eperl -x
6. env -i GATEWAY_INTERFACE=CGI/1.1 PATH_TRANSLATED="$PWD/b.phtml" ./eperl ''
5/6 will have the whole environment on bookworm and trixie.
4 will have the whole environment on bookworm (libperl5.36:i386
5.36.0-7+deb12u2)
but will only have one variable on trixie (libperl5.40:i386 5.40.1-6).
The arguments have no effect on the generated code, environment,
or libperl parameters.
If you apply the patch above
(turning the argument sexion of the environment block
from (2, {"./eperl", NULL}) to (2, {"./eperl", ""}))
4 will also have the whole environment on trixie.
This stinks of UB, but GCC UBsan doesn't pick up anything.
But I haven't been able to reduce this to a minimal reproducer.
I have isolated this to Perl5_Run() doing
// A
/* now allocate the Perl interpreter */
PERL_SYS_INIT3(&myargc, &myargv, &environ);
my_perl = perl_alloc();
perl_construct(my_perl);
// B
/* now parse the script!
NOTICE: At this point, the script gets
only _parsed_, not evaluated/executed! */
rc = perl_parse(my_perl, Perl5_XSInit, myargc, myargv, environ);
// C
where dumping environ at A and B reveals it's full,
but dumping environ at C reveals it's been truncated.
Applying the following, which should, by all accounts, be a no-op,
actually fixes this: environ is undamaged, and the result from the
program is as-expected.
-- >8 --
diff --git a/eperl_perl5.c b/eperl_perl5.c
index 9e66772..a46679a 100644
--- a/eperl_perl5.c
+++ b/eperl_perl5.c
@@ -126,14 +126,14 @@ int Perl5_Run(int myargc, char **myargv, enum
runtime_mode mode, bool fCheck, bo
/* now allocate the Perl interpreter */
- PERL_SYS_INIT3(&myargc, &myargv, &environ);
+ PERL_SYS_INIT(&myargc, &myargv);
my_perl = perl_alloc();
perl_construct(my_perl);
/* now parse the script!
NOTICE: At this point, the script gets
only _parsed_, not evaluated/executed! */
- rc = perl_parse(my_perl, Perl5_XSInit, myargc, myargv, environ);
+ rc = perl_parse(my_perl, Perl5_XSInit, myargc, myargv, NULL);
if (rc != 0) {
IO_restore_stderr();
stdfd_thread_cu(&perlstderr, &stderr_thread);
-- >8 --
This looks like a good patch to include upstream anyway?
but maybe I'm missing something obvious.
Best,
signature.asc
Description: PGP signature
