Remove global array of file descriptors, in
favor of relying on CLO_EXEC exclusively.
Also correctly implement "pipe avoidance" so
that fd redirections do not conflict
with pipes.
This makes two changes to parse trees:
1. Unmaterialized nodes no longer have an invalid source location
For example, with the code `while false;end` there are no tokens
associated with the while loop's job_list, and therefore it is
unmaterialized. Previously it would have had a SOURCE_OFFSET_INVALID.
But now it has a zero source length, but an offset equal to the end of
the while loop (i.e. the semicolon), and a zero length. Correspondingly,
the has_source function now checks the length instead of the offset.
2. Special (comment and error) nodes have always been "disconnected,"
meaning they are not the child of any other node. However, they now have
their parent offsets set to whatever the top of the node stack was when
the node was encountered. This gives us a sense of which node the
comment is "in", e.g. if we are constructing a job list then the
comment's parent will be the job list. This lets us determine the
comment's indent.
If a wildcard or completion expands to a file that begins with
one or more dashes, prepend a ./ to it so that it doesn't get
parsed as an option.
Fixes#1519
The `--null` flag to `read` makes it split incoming lines on NUL instead
of newlines. This is intended for processing the output of a command
that uses NUL separators (such as `find -print0`).
Fixes#1694.
Add the --wraps option to 'complete' and 'function'. This allows a
command to (recursively) inherit the completions of a wrapped command.
Fixes#393.
When evaluating a completion, we inspect the entire "wrap chain" for a
command, i.e. we follow the sequence of wrapping until we either hit a
loop (which we silently ignore) or the end of the chain. We then
evaluate completions as if the wrapping command were substituted with
the wrapped command. Currently this only works for commands, i.e.
'complete --command gco --wraps git\ checkout' won't work (that would
seem to encroaching on abbreviations anyways). It might be useful to
show an error message for that case.
The commandline builtin reflects the commandline with the wrapped
command substituted in, so e.g. git completions (which inspect the
command line) will just work. This sort of command line munging is
also performed by 'complete -C' so it's not totally without precedent.
'alias will also now mark its generated function as wrapping the
'target.
The inotify notifier is fragile, fails on travis, and fails to compile
on certain Linux kernels. It doesn't appear to work as well as the named
pipe mechanism. Best to just get rid of it.
Currently it contains strange code like using `do` loop in order to
avoid `goto`s (they aren't evil, honestly), the pointless `if (mem)`
conditional which doesn't even work (had semicolon for some reason).
You may think this code had a bug where the code didn't check for
the pointer to be null before calling `free`, but this is not the case,
as according to C and C++ standard, `free` should allow `NULL` pointers,
and ignore them.
