Re: `message' not outputting the newline "atomically"

[Paul Eggert]

Eli Zaretskii wrote:

> Bugs aren't hypotheses, they actually happen.

I've looked carefully for any of the bugs hypothesized so far, and haven't found 
any. That's not to say they don't exist, but these mythical creatures are so 
rare that in practice it's fair to assert that they are less important than the 
practical bugs that this patch fixed.

If we could use hypothetical bugs as part of the argument, I could easily 
hypothesize bugs where unbuffered stderr produces less-useful output than 
line-buffered stderr does. But that would be going down a rabbit hole. We should 
focus on bugs that actually occur rather than implausible bugs hypothesized for 
the sake of argument.

> the primary purpose of stderr is to produce diagnostic
> error messages

And if these diagnostics are confusing or unintelligible, as with unbuffered 
stderr, this purpose is not being fulfilled.

> That use case could have a more specialized solution

Sure: we could take every bit of Emacs code that produces piecemeal stderr 
output, and rewrite it to generate the output into a temporary memory buffer and 
then output that buffer manually. For example, we could install something like 
the attached patch to address the dump_fingerprint issue. But this sort of 
change would be counterproductive: it would complicate the code and it would not 
make diagnostic output any more reliable (on the contrary - just look at all 
those tricky buffer-size calculations). In contrast, using _IOLBF is a one-line 
fix that fixes the problem without introducing all this complexity and 
unreliability.

handbuffer.diff
Description: Text Data