Re: WinAPI spawn() not used by Cygwin posix_spawn()? Re: [PATCH] Cygwin: Add spawn family of functions to docs
On Mon, 17 Feb 2025, Corinna Vinschen wrote: > The requirements of posix_spawn and their helper functions are so > that we can't easily fulfill them without doing the fork/exec > twist. > > See https://man7.org/linux/man-pages/man3/posix_spawn.3.html. > Windows CreateProcess() is not quite the same as Linux clone(). > > However, if you think you can come up with a version only running the > spawnve function and thus speed up Cygwin, feel free to send patches. (based on my recollections rather than actually reading the docs again...) Specifically, maybe the case where the attributes which are supposed to do things effectively between fork and exec are not set could be fast-pathed to use spawn/CreateProcess, and the more complex cases fall back to fork/exec. If the created process is also a Cygwin process, perhaps the actions that are supposed to happen between fork and exec could happen in the startup code (much like fork emulation). Just brainstorming. BTW, I vaguely remember an issue with GNU make in MSYS2, where we found that making it *not* use posix_spawn made it faster. https://github.com/msys2/MSYS2-packages/blob/a0902f1e21781022c5ceca44c64190998a62e048/make/PKGBUILD#L31-L32
Re: WinAPI spawn() not used by Cygwin posix_spawn()? Re: [PATCH] Cygwin: Add spawn family of functions to docs
On Mon, 17 Feb 2025 at 11:08, Corinna Vinschen via Cygwin wrote: > > On Feb 16 23:33, Lionel Cons via Cygwin wrote: > > On Sun, 16 Feb 2025 at 22:47, Mark Geisert wrote: > > > > > > In the doc tree, change the title of section "Other UNIX system > > > interfaces..." to "Other system interfaces...". Add the spawn family of > > > functions noting their origin as Windows. > > > > re spawn() family: Cygwin posix_spawn() seems to rely on the rather > > inefficient vfork(), while Opengroup intended it to be an API to > > Windows spawn(). > > > > Is there a technical limitation why Cygwin posix_spawn() cannot use > > WinAPI spawn() directly? > > The requirements of posix_spawn and their helper functions are so > that we can't easily fulfill them without doing the fork/exec > twist. How did UWIN do that? I did ask around - Glenn Fowler of AT&T Research demonstrated a prototype of UWIN posix_spawn() before posix_spawn() was finalised by the Austin Group as Opengroup POSIX standard. So this IS possible, and because non fork() or page cloning has to be done it should be significantly faster than the fork()-and-throw-copied-data-away-at-exec() approach. > See https://man7.org/linux/man-pages/man3/posix_spawn.3.html.> Windows > CreateProcess() is not quite the same as Linux clone(). Here might be the misunderstanding: posix_spawn() is intended NOT to copy anything except the file descriptors and requested attributes, no memory pages and no whatsoever. Everything should be done on the caller's process side, nothing in the child process. Lionel
Re: [PATCH] Cygwin: Add spawn family of functions to docs
On Feb 16 13:46, Mark Geisert wrote: > In the doc tree, change the title of section "Other UNIX system > interfaces..." to "Other system interfaces...". Add the spawn family of > functions noting their origin as Windows. > > The title change seems warranted as neither the spawn family of > functions nor the listed clock_setres() function originated from UNIX > systems. > > --- > winsup/doc/posix.xml | 11 ++- > 1 file changed, 10 insertions(+), 1 deletion(-) Yeah, this is probably the right thing to do. I'm jsut waiting for Brian in terms of the POSIX doc update. That's why I wait with the eaccess and acl functions as well. Thanks, Corinna
Re: WinAPI spawn() not used by Cygwin posix_spawn()? Re: [PATCH] Cygwin: Add spawn family of functions to docs
On Feb 16 23:33, Lionel Cons via Cygwin wrote: > On Sun, 16 Feb 2025 at 22:47, Mark Geisert wrote: > > > > In the doc tree, change the title of section "Other UNIX system > > interfaces..." to "Other system interfaces...". Add the spawn family of > > functions noting their origin as Windows. > > re spawn() family: Cygwin posix_spawn() seems to rely on the rather > inefficient vfork(), while Opengroup intended it to be an API to > Windows spawn(). > > Is there a technical limitation why Cygwin posix_spawn() cannot use > WinAPI spawn() directly? The requirements of posix_spawn and their helper functions are so that we can't easily fulfill them without doing the fork/exec twist. See https://man7.org/linux/man-pages/man3/posix_spawn.3.html. Windows CreateProcess() is not quite the same as Linux clone(). However, if you think you can come up with a version only running the spawnve function and thus speed up Cygwin, feel free to send patches. Corinna
Re: [PATCH] Cygwin: Add spawn family of functions to docs
On 2025-02-17 03:22, Corinna Vinschen wrote: On Feb 16 13:46, Mark Geisert wrote: In the doc tree, change the title of section "Other UNIX system interfaces..." to "Other system interfaces...". Add the spawn family of functions noting their origin as Windows. The title change seems warranted as neither the spawn family of functions nor the listed clock_setres() function originated from UNIX systems. --- winsup/doc/posix.xml | 11 ++- 1 file changed, 10 insertions(+), 1 deletion(-) Yeah, this is probably the right thing to do. I'm jsut waiting for Brian in terms of the POSIX doc update. That's why I wait with the eaccess and acl functions as well. I'll resubmit some time this week. -- Take care. Thanks, Brian Inglis Calgary, Alberta, Canada La perfection est atteinte Perfection is achieved non pas lorsqu'il n'y a plus rien à ajouter not when there is no more to add mais lorsqu'il n'y a plus rien à retrancher but when there is no more to cut -- Antoine de Saint-Exupéry
