Request 1008386 (accepted)

This request supersedes: request 1006370 (Show diff)

Overview

Request 1008386 accepted

- Add upstream patches
* bash52-001
Expanding unset arrays in an arithmetic context can cause a
segmentation fault.
* bash52-002
Starting bash with an invalid locale specification for
LC_ALL/LANG/LC_CTYPE can cause the shell to crash.
- Do not run checks in parallel as it eats memory, a lot of memory
- Disable alternate array implementation as it eats a lot of memory

- Update to final bash 5.2
a. When replacing a history entry, make sure the existing entry has a non-NULL
timestamp before copying it; it may have been added by the application, not
the history library.

- Modernize run-tests

- add checks

- Enable parallel builds by splitting clean and all at make time
- Update to bash 5.2 rc4
Pos. aa is now enabled by default.
m. Readline now checks for changes to locale settings (LC_ALL/LC_CTYPE/LANG)
each time it is called, and modifies the appropriate locale-specific display
- Port patches
* bash-2.03-manual.patch
* bash-5.2.dif

- Update to bash 5.2 rc2
gg. Since there is no `declare -' equivalent of `local -', make sure to use

Created by WernerFink over 1 year ago
In state accepted
Supersedes 1006370

Submit bash

Comments for request 1008386 1
Comments for request 1006370 11

Fabian Vogt (favogt) - over 1 year ago

Fix for tre: https://build.opensuse.org/request/show/1009751

Dr. Werner Fink (WernerFink) - over 1 year ago

author source maintainer

Please use readline from SR#1006359 for testing, otherwise bash 5.2 does not build

Richard Brown (RBrownSUSE) - over 1 year ago

reviewer

Moving this and readline into Staging:L because it seems to break tcsh and tre and parking it in L will leave you the buildlogs while we get the rest of :C moving.

Dr. Werner Fink (WernerFink) - over 1 year ago

author source maintainer

tcsh does not use (lib)readline ... clisp does ... and

NetworkManager
bc
parted
libisoburn1
libgjs0
socat
lua53
lua51
usbprog
perl-TermReadLine-Gnu
x3270
unixODBC
multipath-tools
ntp
ipmitool
OpenIPMI
fontforge
clisp
texlive-asymptote-bin
pipewire-tools
libvirt-client
libvirt-daemon
fvwm2
gnuplot
keepassxc
nftables
hunspell
python
python3
augeas
gpg2
libxml2-tools
util-linux
zypper
bluez
bluez-test
lvm2
samba-client
gdb
ruby2.5-stdlib

Richard Brown (RBrownSUSE) - over 1 year ago

reviewer

Well then if I'm wrong then the outcome of this shuffling of stagings will be bash and readline will get in Factory faster than :C which will still be broken after the rebuild..time will tell, comments will follow :)

Fabian Vogt (favogt) - over 1 year ago

I debugged the tcsh issue. It's because bash 5.2 uses implicit exec in more cases now:

Before:

abuild@fussheizung:~/rpmbuild/BUILD/tcsh-6.24.01> (SHLVL=5 printenv SHLVL)
4
abuild@fussheizung:~/rpmbuild/BUILD/tcsh-6.24.01> (:; SHLVL=5 printenv SHLVL)
5
abuild@fussheizung:~/rpmbuild/BUILD/tcsh-6.24.01> bash --version
GNU bash, version 5.1.16(1)-release (x86_64-suse-linux)

After:

abuild@fussheizung:~/rpmbuild/BUILD/tcsh-6.24.01> (SHLVL=5 printenv SHLVL)
4
abuild@fussheizung:~/rpmbuild/BUILD/tcsh-6.24.01> (:; SHLVL=5 printenv SHLVL)
4
abuild@fussheizung:~/rpmbuild/BUILD/tcsh-6.24.01> bash --version
GNU bash, version 5.2.0(1)-release (x86_64-suse-linux)

Using env to avoid the SHLVL decrement makes it work. The GH fork accepts PRs apparently, so I opened https://github.com/tcsh-org/tcsh/pull/53

Dr. Werner Fink (WernerFink) - over 1 year ago

author source maintainer

See SR#1007204 with patch tcsh-6.24.01-bash52.dif

erner/tcsh> cat tcsh-6.24.01-bash52.dif
--- tcsh-6.24.01/tests/variables.at
+++ tcsh-6.24.01/tests/variables.at     2022-09-30 08:56:23.296533116 +0000
@@ -964,7 +964,7 @@ tcsh -f -c 'tcsh -f -c "printenv SHLVL"'
 tcsh -f -c 'exec tcsh -f -c "printenv SHLVL"'
 tcsh -f -c '(exec tcsh -f -c "printenv SHLVL")'
 ]])
-AT_CHECK([SHLVL=5 tcsh -f shlvl.csh], ,
+AT_CHECK([SHLVL=5 tcsh -f shlvl.csh;:], ,
 [6
 8
 2

Fabian Vogt (favogt) - over 1 year ago

The tre failure is weird. The testsuite is wrong and it should actually fail!

exitstatus.args:

# Some errors which should give exit status 2.
-d .* dummy

But: exitstatus.ok has:

#### TEST: agrep  -d .* dummy exitstatus.in

Exit status 1.

It should actually fail with Exit status 2 because .* should trigger this error:

abuild@fussheizung:~/rpmbuild/BUILD/tre-0.8.0_git201402282055> src/agrep  -d .* dummy exitstatus.in; echo $?
agrep: Record delimiter pattern must not match an empty string
2

Dr. Werner Fink (WernerFink) - over 1 year ago

author source maintainer

Those lines

[   72s] -/usr/src/debug/tre-0.8.0_git201402282055/src/agrep.c 0 (none) 100644 root root 0 4294967295 
[...]
[   72s] +/usr/src/debug/tre-@ VERSION@-@ RELEASE_LONG@ .x86_64/src/agrep.c 0 (none) 100644 root root 0 4294967295

are also fishy

Fabian Vogt (favogt) - over 1 year ago

It's a bug in the test script that it works at all...

The root cause is a behaviour change in bash 5.2. Before:

abuild@fussheizung:~> echo .*
. .. .bash_history .lesshst .rpmmacros .rpmrc .viminfo
abuild@fussheizung:~> bash --version
GNU bash, version 5.1.16(1)-release (x86_64-suse-linux)

With bash 5.2, it no longer matches . and ..!

abuild@fussheizung:~> echo .*
.bash_history .lesshst .rpmmacros .rpmrc .viminfo
abuild@fussheizung:~> bash --version
GNU bash, version 5.2.0(1)-release (x86_64-suse-linux)

The bug in the tre test script is as follows: exitstatus.args contains the line -d .* dummy. This is meant to be passed as three separate arguments to agrep. To achieve that, run-tests.sh does:

for arg in `cat $args`; do
...
$agrep $extra $arg $input >> $out
...

However, this causes glob expansion, so .* is expanded to . ... This totally changes the result... With bash 5.2, there are no matches so .* stays as-is. Question is how to escape this properly so that glob expansion doesn't happen. Or just skip this case, tre is dead anyway.

Dr. Werner Fink (WernerFink) - over 1 year ago

author source maintainer

shopt -u globskipdots does shwitch back

... a line like