> When yum reaches the point of actually installing things, after > all the memory intensive dependancy computations, the only thing it > should need to remember is the ordered list of rpm packages to install > (assuming yum can't just call rpm with the packages in order). It > should be able to free most of the memory it had to use to calculate > the ordered list of packages. I am talking in the order of a few kbytes, > not 300MB. Python is an interpreted language with garbage collection, and address-space fragmentation rears its ugly head. Mere free() won't do enough. A compact() is also needed, but is not available in the API.(!) One way is to build the list of rpm in a file or pipe, and use a shell script for top-level control. Using a file has advantages for divide-and-conquer. Another way to do this is for the top-level to fork(), with the child doing the work for each major phase. Then each child is guaranteed to start out with the same small and compact layout of memory. -- -- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
