Jeff King <peff@xxxxxxxx> writes: >> I think that call should reset line.buf to the original buffer on >> the stack, instead of saying "Ok, I'll ignore the original memory >> not owned by us and instead keep pointing at the allocated memory", >> as the allocation was done as a fallback measure. > > I am not sure I agree. Do we think accessing the stack buffer is somehow > cheaper than the heap buffer (perhaps because of cache effects)? If so, > how much cheaper? This is not about stack vs heap or even "cheaper" (whatever your definition of cheap is). The principle applies equally if the original buffer came from BSS. Perhaps I made it clearer by using a more exaggerated example e.g. a typical expected buffer size of 128 bytes, but the third line of 1000 line input file was an anomaly that is 200k bytes long. I do not want to keep that 200k bytes after finishing to process that third line while using its initial 80 bytes for the remaining 977 lines. By the way, William seemed to be unhappy with die(), but I actually think having a die() in the API may not be a bad thing if the check were about something more sensible. For example, even if a strbuf that can grow dynamically, capping the maximum size and say "Hey this is a one-lne-at-a-time text interface; if we need to grow the buffer to 10MB, there is something wrong and a producer of such an input does not even deserve a nice error message" could be an entirely sensible attitude. But that does not mean an initial allocation should be 10MB. If the expected typical workload fits within a lot lower bound, starting from there and allowing it to grow up to that maximum would be the more natural thing to do. And the problem I have with the proposed "fixed" is that it does not allow us to do that. -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html