http://www.airs.com/blog/archives/55 Ian Lance Taylor Tue, 18 Nov 2008 21:55:07 +0000 http://wordpress.org/?v=1.5.2 http://www.airs.com/blog/archives/55#comment-5028 Tue, 09 Oct 2007 06:14:14 +0000 http://www.airs.com/blog/archives/55#comment-5028 I would be concerned that using hashes would be too slow, since it would force the linker to actually read the input file. Certainly the linker should not only incrementally link newer files; it should incrementally link any file which has changed in any way at all. That is, think of the timestamp as a very high efficiency hash. It would be easy and appropriate to also check that the size hadn't changed, and, on Unix, that the inode was the same. I would be concerned that using hashes would be too slow, since it would force the linker to actually read the input file. Certainly the linker should not only incrementally link newer files; it should incrementally link any file which has changed in any way at all. That is, think of the timestamp as a very high efficiency hash. It would be easy and appropriate to also check that the size hadn’t changed, and, on Unix, that the inode was the same.

]]> http://www.airs.com/blog/archives/55#comment-5021 Tue, 09 Oct 2007 04:12:32 +0000 http://www.airs.com/blog/archives/55#comment-5021 If you ever implement this, please use hashes rather than timestamps (at least as an option). Otherwise there are too many ways to break things; if the .o file is older, but different, the incremental linker still needs to run. Luckily I don't have to use ClearCase anymore, where it was easy for time stamps to move backward. If you ever implement this, please use hashes rather than timestamps (at least as an option). Otherwise there are too many ways to break things; if the .o file is older, but different, the incremental linker still needs to run. Luckily I don’t have to use ClearCase anymore, where it was easy for time stamps to move backward.

]]>