Last modified: 2012-10-14 22:07:43 UTC
The md5sums file for the /latest/ folders contain file names in the format LANG-DATE-XXX, but the files in the folder are named LANG-latest-XXX. This means that one cannot use md5sum -c FILE on that particular file. This is true for all the languages and projects.
I'm not sure that's a bad idea. If the latest folder links redirected to the long form, we wouldn't then find files with latest name which you don't know its timestamp.
The "latest" keyword allows for simple downloading scripts. For my usage, I don't care one bit about the actual date of the dump. I just want to download it, extract and apply it. If the archive is actually the latest, i'm ok with that. However, the current bug is not about the dump folder structure, but about the mismatch between the file names from the folder and the file names from the md5sums file. This need to be fixed one way or the other, in order for the md5sums to be usable.
(In reply to comment #2) > However, the current bug is not about the dump folder structure, but about the > mismatch between the file names from the folder and the file names from the > md5sums file. This need to be fixed one way or the other, in order for the > md5sums to be usable. There is another layer to the mismatch beyond the file naming pattern. If all of the LANG-latest-XXX files were renamed with the LANG-DATE-XXX format of the files they point to, then at least at certain times (early in the month?) we would see a mixture of LANG-DATE_IN_THIS_MONTH-XXX dumps and LANG-DATE_IN_LAST_MONTH-XXX dumps in /latest. However, until all dumps from this month have been completed, LANG-latest-md5sums.txt will point to LANG-DATE_IN_LAST_MONTH-md5sums.txt which will only contain the hashes from last month's dumps. In other words, the xml dumps in /latest "roll over"/update *individually* as soon as they are completed, but the md5sums file only updates when all dumps from that month have been completed.
