Function tables shouldn't be required

[LeoVen]

trafficstars

The library should:

• [ ] Allow fkey and fval parameters to be null
• [ ] Check in _new() for required functions
• [ ] Check if the struct and if the functions are null before every usage and report errors instead of crashing

[duarm]

Fell for this one, I thought f_val was copied into the list, not kept as a reference, so I did something like this

element_new_custom(50, &((struct tile_effects_list_fval){ .cmp = compare_element }), NULL, NULL);

and you can guess what happened

[LeoVen]

Fell for this one, I thought f_val was copied into the list, not kept as a reference, so I did something like this

element_new_custom(50, &((struct tile_effects_list_fval){ .cmp = compare_element }), NULL, NULL);

and you can guess what happened

Now I'm really starting to think if the function tables should be kept as a copy for each ds instead of a pointer

[duarm]

Not sure about the memory / indirect call tradeoff, it seems worth it to just copy the table since it will be most likely on the cache.

on my own modified version, I removed the default allocation table and made it required to be consistent with fval, and one of the problems I noticed is that most libs expect a different table e.g. some expect { .free, .malloc }, while others expect { .realloc, .malloc }, keeping a table for all of them around is annoying. So you might consider copying the allocation table too for consistency