2021年5月4日(火) 22:28 Chet Ramey <chet.ra...@case.edu>: > On 5/4/21 2:31 AM, Koichi Murase wrote: > > (1) Suggestion: use buffered read for a regular file for delim != '\n'? > > > >> It won't work on any system that doesn't return -1/ESPIPE when you try > >> to lseek on a terminal device. > > > > [...] > > > > lseek(2) may succeed on non-seekable streams including some terminal > > device (mentioned in Chet's reply) and a tape device (mentioned in the > > above FreeBSD discussion). > > Let's take a step back. I think we can make the code simpler and still > handle all the cases we need to. I don't think tape devices are worth > worrying about. > > The issue with the BSDs and terminal devices is that the current code > only checks for -1/ESPIPE, but they return a different value for errno > (EINVAL? I forget.) I believe AIX returns something else. Rather than > check for specific errno values, we can just decide not to buffer input > if lseek fails for any reason.
OK. I agree with this direction. The attached file is my current patch for (2) and (3) with the above strategy (no check for ESPIPE). ---- By the way, I'm sorry but there was a typo in my previous mail: > (3) Suggestion: change the order of `run_callback' and `zsyncfd'? > > I noticed that `zsyncfd' is called before `run_callback': This is a typo of "`zsyncfd' is called **after** `run_callback'". -- Koichi
0001-mapfile-buffered-read-in-more-cases-v2.patch
Description: Binary data
