On 4 February 2017 at 21:40, Evgeny Kotkov <evgeny.kot...@visualsvn.com> wrote: > Hi everyone, > > The Unix implementation of apr_file_gets() features an optimization > for buffered files that avoids calling apr_file_read() / memcpy() for > one byte, and directly uses the buffered data. This optimization was > added in https://svn.apache.org/r65294 > > The attached patch implements the same optimization in the Win32 version > of apr_file_gets(), and extends the corresponding part of the test suite. > > This makes the function roughly 4 times faster than before. It is heavily > used by the Subversion's filesystem layer and various modules in the > HTTP Server, e.g., mod_negotiation, and these or other existing callers > would benefit from this optimization. > Hi Evgeny,
Thanks for the patch. Committed in r1785072 [1] PS: apr_file_gets() could be further optimized by using memchr() to find '\n' in the buffer, but I don't know whether performance improvement worth the efforts. https://svn.apache.org/r1785072 -- Ivan Zhakov
