On 12/2/20 12:55 PM, Zebediah Figura (she/her) wrote:
On 12/2/20 11:26 AM, Derek Lesho wrote:
On 12/2/20 12:25 PM, Zebediah Figura (she/her) wrote:
On 12/2/20 11:15 AM, Derek Lesho wrote:
On 12/2/20 12:07 PM, Zebediah Figura (she/her) wrote:
Performance isn't really the concern—it's more code, and you really should be handling allocation failure.
Another solution could be to just use a fixed-size stack buffer.
Okay, I guess I'll do that. There's no limit to how many types you can register for the MFT so I'll just make a max of 100 for both input and output (so ~3.2kb of stack space).
I think there's no need for that; just allocate as much stack space as you actually need.
That varies on a per MFT registration basis, some MFTs require more registration structs than others, since they have more input/output types. Unless you're suggesting I use a VLA?
No, simply use the maximum array size that any individual transform needs.
Not looking forward to the bugs that show up from somebody adding more entries to the supported types list and not increasing the array size just so we can avoid two heap_allocs, but ok.