Return 0 space instead of -EINVAL when there are no readers

[klogg]

Actually no errors are suppressed here. Returning 0 is more correct in this case because -EINVAL would mean there is some invalid value passed on write, which is not true - write request itself is totally fine; returning 0 means no data was successfully written, which is true because there is no reader to consume.

NOTE: this is a split of #26

[stappersg]

We are getting closer.

to be continued

[klogg]

We are getting closer.
to be continued

Didn't get this comment.

[stappersg]

#30 and #31 do obsolete this pull request.

[klogg]

#30 and #31 do obsolete this pull request.

Disagree. You are still returning error which is incorrect in case of 'write'

[daniel-starke]

-ENODEV is also not allowed here. See torvalds/linux@6bfbfcf

[klogg]

Rebased to master.
@jbuchbinder this patch still valid.
@stappersg please see comment from @daniel-starke - changes introduced in #30 and #31 are incorrect from the Linux kernel POV.

[stappersg]

The torvalds/linux@6bfbfcf is https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=6bfbfcfc58005ee12d37b56a5e722618ef6bee8f and says

The tty line disciplines don't expect tty_operations::write_room to return negative values.

Then seeing distinct return values, ENODEV and EINVAL, all being replaced by the same 0 feels wrong.

The

make everyone's write_room return >= 0

also expressed as tty line disciplines don't expect negative values would allow other values as just 0.

Anyway:

• Comment has been seen
• Further research is needed before nullification of an error code
• That research isn't scheduled in a near future

[klogg]

Well, I do not think you need any research here, just following the best practice. It is pretty obvious that kernel drivers express different behavior here, as shown above. Moreover, any existing tty tool will break with negative on write - you can check yourself e.g. with socat.

[jbuchbinder]

Merging to un-break the current behavior with other tools. If we're going to change the return values to be distinct for specific breaking error conditions, that's fine, we can change them to be positive return values. 0 for now, additional patches welcome.