Re: [PATCH] More array filling optimizations

Hi Matt, Sorry, but I'm very busy in the following two weeks (have to travel a lot). I'll try not to forget about you patch, but it would be better if you remember me after June 7. Thanks. Dmitry. Matt Wilmas wrote:

Hi Dmitry, all, I looked into and fixed the -1 size and memory overflow error a couple weeks or so ago, and finally updated the patches now. :-) I made the mistake of assuming the code that checks offset/length (in array_splice, etc.) set length correctly, but it can still be negative, which broke array_init_size... Should be OK now there, and also in array_slice and array_chunk. So if it works correctly now, it sounded like the patch could be used? (This doesn't have the compile/VM part, of course.) I would be able to commit these things myself now, but wouldn't unless I was given the OK first! :-) http://realplain.com/php/array_init_size.diff http://realplain.com/php/array_init_size_5_3.diff Thanks, Matt ----- Original Message ----- From: "Dmitry Stogov" Sent: Wednesday, April 30, 2008

Matt Wilmas wrote:

Hi again Dmitry, Hmm, if a -1 size was passed, I thought it would just make a huge array

size

(bad, obviously), but the tests would still work. (Or I guess it might exhaust memory_limit.) Anyway, the code for checking the size (length variable) in array_splice() is copied from php_splice(), so maybe

there's

some other error if it's becoming -1...

The bug causes memory overflow error, but the size definitely shouldn't be -1, as the same tests work fine before the patch.

You don't think it makes sense to check return_value_used like next/prev/end/reset do? They would only create a 1 element array, but array_splice() could waste time filling a large array that isn't used.

I

think the function is used a lot to only remove/replace elements, not

using

the return_value: http://www.google.com/codesearch?q=lang%3Aphp+array_splice

Ah. I see. My mistake. Of course it makes sense. Thanks. Dmitry.

If there's a -1 size bug, the array_splice changes can be reverted and simply have: if (return_value_used) {

    /* Initialize return value */
    array_init(return_value);
    rem_hash = &Z_ARRVAL_P(return_value);

} - Matt ----- Original Message ----- From: "Dmitry Stogov" Sent: Wednesday, April 30, 2008

For some reason "make test" with the patch reported several broken array_splice() tests. Looking in gdb I saw that init_array() got -1 as

a

size of new array. I don't think checks for return_value_used for array_splice() have a

lot

of sense. Thanks. Dmitry.