Bert Huijben wrote: > Julian Foad wrote: >>>> URL: http://svn.apache.org/viewvc?rev=1344347&view=rev >> This commit implements buffering of a generic stream, directly in the config >> file parser. Would it not be better to implement buffering of a generic >> stream >> as a generic module? > > Perhaps... > > But a patch like that didn't allow speeding up the test suite by 15% in one > afternoon. This was a very local patch with a very huge result. (And as it > was very local it is easy to remove it when somebody else optimizes it in a > different place).
Sure -- I appreciate why it was worth doing that way. > [...] > > And just abstracting it, just for abstracting will add more overhead... (The > current code is all in one C file and will be completely inlined by every > compiler... Which is impossible if we need 3 callback and/or shared library > calls for reading every byte... Even worse when that is taking out mutexes > such as the apr read code) I don't think it's fair to assume that an abstraction need be that inefficient, although I admit that an implementation easily can be that inefficient if we don't take care of it. Part of what I was thinking is, why are we re-inventing the standard <stdio.h> streams interface (fread, fgetc, fungetc, ...) in Subversion? And I started wondering if we could hook in to the standard streams. The GNU C library (glibc) does provide hooks for doing this: http://www.gnu.org/software/libc/manual/html_mono/libc.html#Custom-Streams But it's not standardized, and Windows doesn't seem to provide any such facility. I don't think we need to do anything right now, but I wanted to note the suggestion. - Julian
