Ramsay Jones <ram...@ramsayjones.plus.com> writes: > In a previous comment, I said that (on 32-bit Linux) it was likely > that an object of > 4GB could not be handled correctly anyway. (more > likely > 2GB). This was based on the code from (quite some) years ago. > In particular, before you added the "streaming API". So, maybe a 32-bit > arch _should_ be able to handle objects as large as the LFS API allows. > (Ignoring, for the moment, that I think anybody who puts files of that > size into an SCM probably gets what they deserve. :-P ). > > The two patches I commented on, however, changed the type of some > variables from off_t to size_t. In general, the patches did not > seem to make anything worse, but these type changes could potentially > do harm. Hence my comment. (I still haven't tried the patches on my > 32-bit Linux system. I only boot it up about once a week, and I would > rather wait until the patches are in the 'pu' branch before testing).
We are in perfect agreement. I didn't mean to say that it is OK to replace off_t with size_t without a good reason, especially when the current code (at least the part I looked at anyway, like the OFS_DELTA part) seems to use off_t correctly, and your review comments are very much appreciated, so is the effort started by Martin to take us in the direction of using types more appropriate than "ulong".
