On 28.04.2019 01:34, Joerg Sonnenberger wrote: > On Fri, Apr 26, 2019 at 11:36:00PM +0200, Kamil Rytarowski wrote: >> I propose a patch that: > > Too many different things at once. So the short version of the review: > - I don't see any reason for providing unlocked versions of clearerr, > feof, ferror or fileno. No reason for inline functions either, they > shouldn't be performance sensitive on any reasonable code. There are still potential users, but most of them in open source packages are from copy-pasted GNU scripts. One legitimate user of fileno_unlocked (from a feature point of view) is in TSan, but it has been reimplemented inside the sanitizer for the NetBSD case. > - getc is performance sensitive. putc potentially is, but I don't have > clear evidence for the relevant use. Just dropping the inline > optimisation for getc results in a penalty between 3 and 10, depending > on the architecture, for single-threaded programs. That's clearly not > acceptable. (Conditionally) switching from getc macro to > getc_unlocked would be reasonable on platforms like AMD64, but still > include a 80% overhead (for static linking!) on HPPA. > OK, so if its considered as performance sensitive I will abandon that part and enforce _REENTRANT in GCC too. As we will keep using inline macros, no point in switching feof(3) and friends. > Joerg > OK, thanks!
Attachment:
signature.asc
Description: OpenPGP digital signature