corec: rework internals to avoid pointer alignment issues #204

robUx4 · 2025-01-03T17:10:01Z

void* is fine if we cast to the proper target type.
It's not fully clean as the structure as the array allocations don't take care of the final pointer alignments.

The address sanitizer is happy and the regression tests still pass.

Use ARRAYCOUNT when you know the type inside the array.

We can just share the static structure with node.c.

With stronger typing.

And only reset _Begin/_End once.

Using direct access to the array fields

Just count the offset, to void some pointer arithmetics.

So we are free to use the alignment we want. And we don't do pointer arithmetics.

So we don't deal with alignment which is correct since it's pointing to an actual pointer of that type.

It's potentially using less space and needs less pointer subtractions.

robUx4 added 30 commits January 3, 2025 18:13

libebml2: fix use after free on CRC reading error

a2ae032

libebml2: pass the CRC data as void

7488c94

libmatroska2: count block data with ARRAYCOUNT()

bec2a57

corec: make the string tab type fully private

ab275ce

corec: make ArraySize() private

1f79410

Use ARRAYCOUNT when you know the type inside the array.

corec: make DATA_FLAG_HEAP private

aad0993

corec: make DATA_FLAG_MEMHEAP private

99df83a

We can just share the static structure with node.c.

corec: use ArrayInitEx() to clear memheap arrays

c4189e8

corec: expand Data_Release() into Data_Clear()

b9f4076

corec: reorder Data_Clear()

43b64fb

corec: turn memheap macros to inline functions

577c421

With stronger typing.

corec: turn memheap macros to inline functions

c496222

With stronger typing.

corec: make MemHeap_Null() private to the array

192b678

corec: use a special macro to get the cc_memheap

2ab509a

corec: stronger typing of dataheaphead macros

da88e2f

corec: stronger typing of datahead macros

da883d1

corec: cast early to datahead

8bdaa7c

corec: simplify access to dataheaphead

7556bd8

corec: pass the array directly to Data_Clear()

ccd0f6b

corec: merge Data_Clear() and ArrayClear()

c081de1

And only reset _Begin/_End once.

corec: pass the array directly to Data_ReAlloc()

06644bc

corec: simplify Data_ReAlloc()

743532a

Using direct access to the array fields

corec: simplify datahead realloc code

5358c19

corec: only get datahead from an array

d474e0e

corec: don't use Data_Head() on NULL arrays

3ac1437

corec: don't use a moving pointer in the FIFO

3002db6

Just count the offset, to void some pointer arithmetics.

corec: use void* for array pointers

cbab81a

So we are free to use the alignment we want. And we don't do pointer arithmetics.

corec: cast internal node pointers using void*

bc4394e

So we don't deal with alignment which is correct since it's pointing to an actual pointer of that type.

corec: store the used size rather than the end pointer

80859ca

It's potentially using less space and needs less pointer subtractions.

corec: remove unused macro

40b7862

robUx4 added 2 commits January 3, 2025 18:13

corec: only get dataheaphead from an array

235c9bb

corec: inline MemHeap_Null()

13026f4

robUx4 force-pushed the alignfux branch from d9183dc to 13026f4 Compare January 3, 2025 17:14

robUx4 merged commit cb01450 into Matroska-Org:master Jan 3, 2025
5 checks passed

robUx4 deleted the alignfux branch January 3, 2025 17:16

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

corec: rework internals to avoid pointer alignment issues #204

corec: rework internals to avoid pointer alignment issues #204

robUx4 commented Jan 3, 2025 •

edited

Loading

corec: rework internals to avoid pointer alignment issues #204

corec: rework internals to avoid pointer alignment issues #204

Conversation

robUx4 commented Jan 3, 2025 • edited Loading

robUx4 commented Jan 3, 2025 •

edited

Loading