> From: Stephen Hemminger [mailto:step...@networkplumber.org] > Sent: Thursday, 9 November 2023 16.45 > > On Thu, 9 Nov 2023 09:21:22 +0100 > Morten Brørup <m...@smartsharesystems.com> wrote: > > > I can't find the definition of BUFSIZ. Please make sure to add a > comment to the definition of BUFSIZ mentioning - like in your patch > description - that it will be more than sufficient for the info related > blocks in the file. > > > > More comments inline below, regarding existing bugs found while > reviewing. > > > > > > Assuming BUFSIZ has a comment describing the reason for its value, > > > > Acked-by: Morten Brørup <m...@smartsharesystems.com> > > The constant BUFSIZ comes from stdio.h and used lots of places in > libraries. > It is 8192 in current glibc and unlikely to be a problem.
OK, didn't know that. So I looked it up, trying to learn more about it. I found two sources [1], [2] mentioning that BUFSIZ is guaranteed to be at least 256. [1]: https://www.gnu.org/software/libc/manual/html_node/Controlling-Buffering.html#BUFSIZ [2]: Page 234 in "The C Standard Library" by P.J. Plauger, ISBN: 0-13-131509-9, from 1992 If 256 suffices, then I am OK with using BUFSIZ. I hope the authors of the other libraries using BUFSIZ don't assume more than the C standard promises about it. > Chose it because this a on stack buffer used before writing to a file > which > is similar to what stdio does. > > The library does not use stdio because most of the I/O is writing > packets > which needs to be fast and overhead of extra stdio buffer is harmful. > Looking into using io_uring in a future version.
