| Age | Commit message (Collapse) | Author |
|---|
|
Doesn't quite work yet though.
|
|
|
|
Just re-open them to clear EOF.
|
|
O_RDWR is a dirty hack to get around the issue of infinite
EOFs while reading an in-FIFO.
Instead, stop breaking POSIX and set the FIFOs to O_RDONLY.
In case a read returns EOF (r == 0), we reopen the fd.
Same will be applied to the friend-fifos (especially file_in),
helping us get rid of strange timeouts and heuristics and
rather solve the problem the POSIX-way.
The only downside to this is that we are blind for writes
to the in-FIFOs between catching read == 0 and close(), but this is
not an issue.
To make reopening as easy as possible, I added a dirfd
to all slots.
While at it, I changed the initial setup and removed the chdir()
in favor of the POSIX-2008-compliant *at-functions.
This lets us do stuff without having to use snprintf to build
paths and is more bulletproof even in case the directory is
renamed.
|
|
|
|
|
|
or FOLDER
So use STATIC here.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
For aborted transfers, we will have to move the transfer to the
TRANSFER_KILLED state. Then we'll have to wait until the input
FIFO stops being readable (the user ^C the transfer). Only then can
we move to TRANSFER_NONE.
If we move to TRANSFER_NONE immediately then we will initiate a
new transfer which we don't want to do.
The basic idea is:
If we move into the TRANSFER_KILLED state and timeout has expired (50ms?)
without the fd being readable (the user did a ^C on cat file > file_in)
then move the transfer to TRANSFER_NONE state.
|
|
|
|
|
