Eric Blake wrote: > The libvirt project would be very interested in using open_memstream,
And many other people, I'm sure, once they learn that it exists and can be used portably. > since it is now part of POSIX 2008. We even tossed around an idea > on IRC on how to implement it for mingw, details below. > > For now, I've got a test working on Linux and cygwin 1.7 as the golden > reference (two independent implementations both satisfying the example > from POSIX 2008), then for BSD systems with funopen(), a light-weight > implementation that does all the work using stdio hooking (borrowing > heavily from my cygwin implementation of open_memstream). I'm posting > the first two patches now, for any feedback, although the third patch > for a heavy-weight implementation for Solaris, cygwin 1.5.x, and mingw > will not be any sooner than Monday. > > For platforms without stdio hooking, the simplest thing we could think > of is to create a temporary file under the hood, then provide gnulib > overrides of fflush and fclose (the only two points at which POSIX > requires that the original arguments to open_memstream are in sync > with what has happened to the stream). Those hooks would then dump > the current contents of the temp file into malloc'd memory. So it is > I/O intensive, but seems feasible to implement. Feasible indeed. We don't care much about efficiency on deficient systems, so this should be fine. Eventually, they'll either catch up or die off. > Other alternatives we > considered were using mmap of a temporary file, and doing a gnulib > override of malloc/realloc to recognize the mmap pointer, which is > less I/O intensive but still requires the disk; but that runs into > mmap difficulties (and makes mingw porting harder). We also > considered hooking every single function that takes a FILE* to > understand string streams, but that was ruled out as too invasive to > be practical. Nice work. Thanks!
