On Thu, 16 Dec 1999, Fred L. Drake, Jr. wrote:

[Note that Greg's message went to python-checkins since he responded to a checkin message, but I suspect he meant to change the header to point to python-dev. ;) If not, too bad!]

I didn't really care too much where it went. I would actually suggest that the Reply-To: on the checkin list is set to python-dev if that is where replies are Supposed To Go. [ I do this with mod_dav checkins; replies to dav-checkins mail goes to dav-dev. ]

Greg Stein writes:

...

But this means that your tables no long reside in "const" space. Yet More Per-Process Memory...

It would be nice to have those tables marked as "const".

Perhaps; as Guido points out, there haven't been a lot of complaints about this issue. I will note that only the tables aren't constant; the strings that are pointed to are still constant. I'm inclined to let the compiler/ linker care about this, and not change the code without a really clear need to do so. Here are the sizes of those tables and the strings they point to (including terminating null bytes for the strings):

pathconf_names: 14 entries, 112 bytes, 176 string bytes confstr_names: 25 entries, 200 bytes, 576 string bytes sysconf_names: 108 entries, 864 bytes, 1774 string bytes

Figures are for Solaris7.

Ah. I just replied to that. Guess that one went to python-checkins :-) True, this is a small amount of memory. But they start to add up. non-const globals also pain me when I start to work on free-threading stuff (each must be examined to see if synchronization is needed), so reducing the number there is important. Regarding the memory itself: as I mentioned in the other note, I just want to ensure that Python's working set remains low (reasons given in that email). Cheers, -g -- Greg Stein, http://www.lyra.org/

On Thu, 16 Dec 1999, Fred L. Drake, Jr. wrote:

[Note that Greg's message went to python-checkins since he responded to a checkin message, but I suspect he meant to change the header to point to python-dev. ;) If not, too bad!]

I didn't really care too much where it went. I would actually suggest that the Reply-To: on the checkin list is set to python-dev if that is where replies are Supposed To Go. [ I do this with mod_dav checkins; replies to dav-checkins mail goes to dav-dev. ]

Greg Stein writes:

...

But this means that your tables no long reside in "const" space. Yet More Per-Process Memory...

It would be nice to have those tables marked as "const".

Perhaps; as Guido points out, there haven't been a lot of complaints about this issue. I will note that only the tables aren't constant; the strings that are pointed to are still constant. I'm inclined to let the compiler/ linker care about this, and not change the code without a really clear need to do so. Here are the sizes of those tables and the strings they point to (including terminating null bytes for the strings):

pathconf_names: 14 entries, 112 bytes, 176 string bytes confstr_names: 25 entries, 200 bytes, 576 string bytes sysconf_names: 108 entries, 864 bytes, 1774 string bytes

Figures are for Solaris7.

Ah. I just replied to that. Guess that one went to python-checkins :-) True, this is a small amount of memory. But they start to add up. non-const globals also pain me when I start to work on free-threading stuff (each must be examined to see if synchronization is needed), so reducing the number there is important. Regarding the memory itself: as I mentioned in the other note, I just want to ensure that Python's working set remains low (reasons given in that email). Cheers, -g -- Greg Stein, http://www.lyra.org/