On 18/11/2026 22:35, Collin Funk wrote:
Pádraig Brady <[email protected]> writes:
On 18/11/2026 21:37, Paul Eggert wrote:
On 2026-11-18 07:30, Martin Mirchev wrote:
POSIX allows read() to
return fewer bytes than requested even when more data follows
Not for regular files. Unless a signal arrives, or the read buffer
size
exceeds SSIZE_MAX, or the resulting file offset would exceed the maximum
off_t value (none of which is the case here), POSIX doesn't allow short
reads of regular files that have more data following.
this
happens for instance on FUSE files opened with direct I/O
That behavior of FUSE breaks applications like dd, and there is no
reasonable workaround there. But I'm not surprised about this, as FUSE
fails to conform to POSIX in other ways, ways that break Git etc.
If the workaround for 'head' is easy then we might as well do it.
But
please let's not say that we're fixing a 'head' bug or because because
POSIX allows short reads of regular files: it's because we're working
around a failure of FUSE to conform to POSIX.
Right agreed.
The workaround is to use full_read() so simple enough.
I'll ensure to say these are "edge cases" in NEWS,
and might not even mention this in NEWS at all.
Marking this bug as done.
Doesn't full_read also retry on EINTR because it uses safe_read?
Not sure if the behavior change really matters, but figured it was worth
mentioning.
Right, though neither head(1) nor tail(1)
define a returning signal handler,
so I don't see any practical unintended behavior change here.
cheers,
Padraig