Rasmus Lerdorf:
[ Charset ISO-8859-1 unsupported, converting... ]
> On 6/23/10 6:56 AM, Wietse Venema wrote:
> > Stas Malyshev:
> >> Hi!
> >>
> >>> could anybody tell me why also for every single php file engine must lstat
> >>> all path?
> >>> Why php engine don't simply try to open directly the file?
> >>
> >> There are some places where PHP engine has to know "true" name of the
> >> file - i.e. filename that would be the same for each file regardless of
> >> how it is accessed - through relative path, symlinks, etc. Example can
> >> be {include|require}_once. For that, it needs to figure out the whole
> >> path. It is done only once per file, then cached. If you seeing it on
> >> repeated runs in the same process, increase your realpath cache size
> >> (yes, the default is way too small for any big app).
> >
> > Another option may be to compute a hash of the file. When you find
> > that file content has the same hash value as a known file you flag
> > them as "same file". This could be faster (file contents are cached
> > in main memory because the OS already knows the files are the same,
> > and finding the "real" name of a file is a slow process).
>
> When properly configured the realpath result should be cached as well
> though and we avoid calling a hash function on what could potentially be
> a lot of data.
The idea is to replace lstat system calls by hashing; in both cases
one would cache the result. Which is faster depends on the number
of pathname elements: system calls are relatively expensive, and
memory mapping could avoid the need for explicit file read() system
calls. If someone has the time then they can test if my idea is
totally bonkers (which it may well be).
Wietse
--
PHP Internals - PHP Runtime Development Mailing List
To unsubscribe, visit: http://www.php.net/unsub.php
