in following simple code, load 1-channel data texture. use glteximage2d() gl_luminance (which 1-channel format) , gl_unsigned_byte, should take 1 byte per pixel. allocate buffer size equal number of pixels (2 x 2) represents input pixel data (the values of pixels don't matter our purposes).
when run following code address sanitizer enabled, detects heap buffer overflow in call glteximage2d(), saying tried read beyond bounds of heap-allocated buffer:
#import <opengles/es2/gl.h> //... eaglcontext* context = [[eaglcontext alloc] initwithapi:keaglrenderingapiopengles2]; [eaglcontext setcurrentcontext:context]; glsizei width = 2, height = 2; void *data = malloc(width * height); // contents don't matter glteximage2d(gl_texture_2d, 0, gl_luminance, width, height, 0, gl_luminance, gl_unsigned_byte, data); this 100% reproducible , happens on both ios simulator , device. if increase size of buffer 6 not overflow (2 bigger expected size of 4).
sizes of 1x1 , 4x4 don't seem have problem, 2x2 , 3x3 do. seems kind of arbitrary.
what wrong?
i have solved @genpfault's comment.
i need set unpack alignment 1:
glpixelstorei(gl_unpack_alignment, 1); specifically, unpack alignment determines alignment start of each row. default value 4. since rows don't have special alignment, , there no gaps between row bytes, alignment should 1.
the first row aligned because malloc allocates 16-aligned buffers. second , subsequent rows misaligned default alignment of 4 unless row length multiple of 4 (this explains why 2x2 , 3x3 don't work, 4x4 does). 1x1 happens work because has no second row.
