> From: "Dave Murphy" <wintermute2k4@...>
> Sent: 2005 October 04 Tuesday 19:19
> Does anyone know if there's a list of Sourceforge mirrors somewhere
> other than the download pages? If it's possible I think it might be
> worth looking into some dynamic regeneration of the mirror lists.
Well, just some ideas.
The download page always seems to list 19 random sites. I think it's
the same format for any download. Is there no way to process this with
a net installer? You could write a script to process it, spit out the
list of names, run X iterations (like 5 or 10), storing only unique
names, and then ship this list with the installer, just to get them
started (or incase the next step fails). Perhaps on the MinGW site, you
could hit (i.e. GET request) this script with an option to send the file
back, thus let it be user-driven iterations. Periodically (cron, or
count time since last hit) check the list for dead entries and remove
Process by looking for "<A HREF=/index-sf.html?use_mirror=server>", and
the form is always server.prdownloads.sourceforge.net.
>>9. A UI for selecting a mix of Current/Proposed would be lovely :)
> Urk, that might get a bit complicated. I'm not even sure how such a
> system would work.
Well, first, requires a parsed FRS (or manually updated) to get the
categories and such. For the UI, perhaps a wider column on the left for
file name, and on the right, columns for each category, with check
boxes. Greyed out if there is nothing, checkable if there is something.
For the logic to make sure two versions do not collide (gcc), I defer to
>>12. With all tarballs already downloaded, and barely sufficient space
>>on the drive, the progress log showed errors installing cc1plus.exe (I
>>think); the error repeated a number of times, and then the processing
>>continued to the next screen, reporting no errors. To fix this, I
>>needed to uninstall.
> Bad assumption on my part here I think. I assumed that the installer
> would check the disk space requirements before proceeding but I've
> since realised that it's dependent on the disk cluster size, number of
> files etc. I've added a check for errors in extraction but I'm not
> entirely sure what to do in the case of failure. For a first install
> it would probably be reasonable to remove the files installed so far
> but when updating it seems a bit harsh to remove the entire
> installation. Having said that, the chances are that a failed
> extraction will have broken the installed toolchain anyway so perhaps
> it's better to be safe than sorry.
Probably stop on disc full, look at what package you have, and how far
it got, notify user through some color or text status message, roll back
the current unsuccessful tool, report what succeeded and what failed,
and how much space would be needed to complete. If it was not being
updated, don't uninstall it. If it was not installed, notify user.
Thanks again for the efforts!