merge from gcc

This commit is contained in:
DJ Delorie
2006-04-12 18:42:01 +00:00
parent 01d3a6ce13
commit 3db2e6ddf6
11 changed files with 387 additions and 51 deletions

View File

@@ -1,3 +1,4 @@
@c -*- mode: texinfo -*-
@deftypefn Extension {struct pex_obj *} pex_init (int @var{flags}, const char *@var{pname}, const char *@var{tempbase})
Prepare to execute one or more programs, with standard output of each
@@ -129,6 +130,67 @@ value, or to 0 if there is no relevant @code{errno}.
@end deftypefn
@deftypefn Extension {FILE *} pex_input_file (struct pex_obj *@var{obj}, int @var{flags}, const char *@var{in_name})
Return a stream for a temporary file to pass to the first program in
the pipeline as input.
The name of the input file is chosen according to the same rules
@code{pex_run} uses to choose output file names, based on
@var{in_name}, @var{obj} and the @code{PEX_SUFFIX} bit in @var{flags}.
Don't call @code{fclose} on the returned stream; the first call to
@code{pex_run} closes it automatically.
If @var{flags} includes @code{PEX_BINARY_OUTPUT}, open the stream in
binary mode; otherwise, open it in the default mode. Including
@code{PEX_BINARY_OUTPUT} in @var{flags} has no effect on Unix.
@end deftypefn
@deftypefn Extension {FILE *} pex_input_pipe (struct pex_obj *@var{obj}, int @var{binary})
Return a stream @var{fp} for a pipe connected to the standard input of
the first program in the pipeline; @var{fp} is opened for writing.
You must have passed @code{PEX_USE_PIPES} to the @code{pex_init} call
that returned @var{obj}.
You must close @var{fp} using @code{fclose} yourself when you have
finished writing data to the pipeline.
The file descriptor underlying @var{fp} is marked not to be inherited
by child processes.
On systems that do not support pipes, this function returns
@code{NULL}, and sets @code{errno} to @code{EINVAL}. If you would
like to write code that is portable to all systems the @code{pex}
functions support, consider using @code{pex_input_file} instead.
There are two opportunities for deadlock using
@code{pex_input_pipe}:
@itemize @bullet
@item
Most systems' pipes can buffer only a fixed amount of data; a process
that writes to a full pipe blocks. Thus, if you write to @file{fp}
before starting the first process, you run the risk of blocking when
there is no child process yet to read the data and allow you to
continue. @code{pex_input_pipe} makes no promises about the
size of the pipe's buffer, so if you need to write any data at all
before starting the first process in the pipeline, consider using
@code{pex_input_file} instead.
@item
Using @code{pex_input_pipe} and @code{pex_read_output} together
may also cause deadlock. If the output pipe fills up, so that each
program in the pipeline is waiting for the next to read more data, and
you fill the input pipe by writing more data to @var{fp}, then there
is no way to make progress: the only process that could read data from
the output pipe is you, but you are blocked on the input pipe.
@end itemize
@end deftypefn
@deftypefn Extension {FILE *} pex_read_output (struct pex_obj *@var{obj}, int @var{binary})
Returns a @code{FILE} pointer which may be used to read the standard