Wayne wrote: >On Tue, 24 Sep 2002, Bill Bogstad wrote: >> If you don't think this should be happening, then I can run trn under >> gdb again and try to figure out the cirumstances when I see this >> happening. > >I'm pretty stale on trn code these days, but obviously the code had >assumed that the list would never grow lower than the lowest value set >during the list's creation. Of all the calls to new_list(), only one of >them passes in something besides "0" as the low value, so I assume that >your crash case is concerning the "article_list", which passes in a low >value of "absfirst". Theoretically, no article should exist lower than >this value, but I don't remember with any certainty if the code ever >guesses this value in some cases and then resets absfirst value downward >later on. If such is the case, I'd prefer to have the code tweak the >list->low value when it modifies the absfirst value downward (rather >than adding code for this to list.c). I looked around in the code and I'ld consider that a breaking of the abstraction that the LIST type provides to the rest of trn's code. Truthfully, I also can't follow the listnum2listitem() code well enough to be sure of the effects if the code actually tried to access something in the new range of items. One concern, I have is that depending on access patterns; we could end up with two LISTNODES with overlaping ranges in the article_list. I think that wherever we change list->low, we are going to have to reduce it by some multiple of list->items_per_node and not just set it to the new (lower) absfirst value. >didn't see any place that might reset absfirst downward, so perhaps you >could let me know in what circumstances this crash can occur? Is the >list in question actually the article_list? If so, is the asked-for >value also lower than absfirst? The only time I saw a crash was when it was the article_list list. I wasn't aware of absfirst, so I don't know anything about its value at the time. I'll see if I can provoke the problem again and get more info... Bill Bogstad bogstad@pobox.com ------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf
