From: Godmar Back (gback@cs.utah.edu)
Date: Tue Dec 08 1998 - 13:02:05 EST
This is a point of discussion that I want to bring up, but it's not
critical to this release. It concerns Edouard's recent additions to jar.c
The current configuration tests whether mmap() is available, and if so,
jar.c uses it and expects it to work. This is okay on most architectures.
However, there are ports where
a) The way in which jar.c uses mmap does not save you much because read()
does not involve a system call. That is, jar_read from the mmapped
file has the same cost as doing the actual read().
and/or
b) Some files are mmappable, others are not. For instance, a file that
resides on a RAM disk might be mmappable, but a file sitting in resident
storage might not be. In these cases, it would be nice if jar.c would
fall back to read() if mmap fails.
- Godmar
This archive was generated by hypermail 2b29 : Sat Sep 23 2000 - 19:57:08 EDT