920384 - assertNoOverlap should allow zero-length regions

Status: RESOLVED DUPLICATE of bug 917593

(Core :: MFBT, defect)

Description

[Steve Fink [:sfink] [:s:]]

I made a modification to the structured clone algorithm that resulted in an assertion failure while trying to deserialize an empty string.

The problem is that the previous allocation at that time was the clone buffer data itself, and I malloc'd 2 bytes (for a NUL that I supposedly might need) to hold the string data immediately after. I then tried to copy 0 bytes of data from end of the clone buffer to the beginning of my malloced buffer, and both of those were at the same pointer location. As a result, I ended up calling copyAndSwapFromLittleEndian with nelems=0. Endian.h's assertNoOverlap got very unhappy about this.

Comment 1

[Steve Fink [:sfink] [:s:]]

Attachment: assertNoOverlap should allow zero-length regions

assertNoOverlap seems unnecessarily complicated, too, since the count is the same for both buffers and so you don't need to worry about the tricky cases. I switched to saying "the distance between the pointers must be at least the count". It uses abs(), though, and maybe that's verboten.

Comment 2

[:Ms2ger (he/him; ⌚ UTC+1/+2)]

How is this different from bug 917593?