buffer.toString() fails when buffer.length >= 32 and null-terminated on big-endian

Version

v24.18.0, v26.5.0

Platform

Linux zelenka 6.12.96+deb13-s390x #1 SMP Debian 6.12.96-1 (2026-07-20) s390x GNU/Linux

Subsystem

StringBytes

What steps will reproduce the bug?

Run these commands

node -p 'const b=Buffer.alloc(32); b.write("\u3000",0,"utf8");b.toString().charCodeAt(0);'
node -p 'Buffer.from("\u3000").toString().charCodeAt(0);'

How often does it reproduce? Is there a required condition?

Always reproduced on big-endian.

What is the expected behavior? Why is that the expected behavior?

both commands should return 12288

What do you see instead?

On big-endian architecture (e.g. s390x) the first call returns 0 and the last 12288.

Additional information

Tested with the executables provided by nodejs.org.
v22.23.1 is not affected.
Also, this gives the correct result on big-endian:

node -p 'const b=Buffer.alloc(31); b.write("\u3000",0,"utf8");b.toString().charCodeAt(0);'

So the suspected commits are the ones adding the special cases for buflen >= 32 here:
https://github.com/nodejs/node/blob/68130800586634f17812f3b0837d869a36725e98/src/string_bytes.cc#L751
and here
https://github.com/nodejs/node/blob/68130800586634f17812f3b0837d869a36725e98/src/string_bytes.cc#L625

Exécutez les deux commandes de reproduction sur un système big-endian et comparez-les au cas de 31 octets. Inspectez ensuite src/string_bytes.cc autour des lignes référencées 625 et 751, en vous concentrant sur le traitement spécial de buflen >= 32. C’est terminé lorsque les deux commandes renvoient 12288 sur big-endian sans régression du comportement existant pour 31 octets.