gh-139184: set O_CLOEXEC for master_fd when call os.forkpty

[Zheaoli]

• Issue: os.forkpty() and pty.fork() return FD with O_CLOEXEC #139184

[Zheaoli]

@vstinner Thanks for your time. PTAL when you get time

[Zheaoli]

@vstinner Done

[picnixz]

os.* functions are meant to mimic libc ones. As such, I'm not sure we should forcibly change inheritance (or does forkpty actually sets O_CLOEXEC as well?). Instead I would suggest one of the two:

• add an additional parameter to forkpty to allow setting O_CLOEXEC at the same time.
• add a separate function (I don't have a name for it) that does it.

[picnixz]

I don't think you need this change.

[vstinner]

I like this change, it helps readability :-)

[picnixz]

Considering the function is already very compact, I don't think it helps much. But I wouldn't mind having more blank lines in the entire function to really help readability (the return is usually sufficiently clear IMO; I'm more worried about the all those ifs before actually calling forkpty)

[vstinner]

os.* functions are meant to mimic libc ones. As such, I'm not sure we should forcibly change inheritance

There is PEP 446 "Make newly created file descriptors non-inheritable" which requires Python to create new file descriptors as non-inheritable.

I see this change as a bugfix, but I prefer to not backport it to avoid bad surprises to users.

[vstinner]

LGTM. Thanks for the update! It now looks good to me with PyErr_FormatUnraisable().

[vstinner]

Merged, thank you.

[Zheaoli]

Merged, thank you.

@vstinner @picnixz thanks for your time!