My project directory was erased

Somehow your build_root was the same as your project directory. Don't know why, but here's a patch which makes the check, that $build_root is set to the current directory more robust (instead of only checking $build_root != ".").

Yes, that's right, $build_root was pointing at my project dir (I was having problems getting makepackage to find any files at all, so I tried that).

Spent two full days on grepping my /dev/sdb1 (extracting all strings to giant files on a separate disk). Then searching for fragments of known files... The missing source dir is back in place again. Sigh of relief.

I applied the path and on running it I get:

The meta package is 'fsect 1.0.package.meta'.
The complete package is 'fsect 1.0.package'.
The Luau repository XML entry file is 'myproject.xml'.
/usr/bin/makepackage: line 267: cd: : No such file or directory
/usr/bin/makepackage: line 267: cd: : No such file or directory

$build_root is set to '.' from the the spec file. It's empty here?

Regards

// ATS.

Hm... Don't know why.

I think as all these directories are starting with apkg- (and are therefore deleted anyway), cleanUp shouldn't explicit delete them (right grammar?).

Here's a new patch which simply doesn't delete the directories, but there shouldn't be any left-overs nevertheless.

Ouch. Nasty thing. Glad you got your stuff back!

We should definitely check more rigorously before cleaning up stuff. The problem is, we can't guard for every possible setting of $build_root (and other variables) I mean, setting $build_root to /usr or /home would be devastating.

I'm gonna look into applying erase.diff into svn. That's on check we should do at the very least.

erase.diff applied in r2442. Can we close this as fixed?

Hello,

What was needed to trigger this was to set both of $build_root and $source_dir to point to the same directory (like /home/ats/MyProject).

I haven't tested the patch but just setting the variables as above would trigger it.

Wouldn't it be possible to store the name of any created directory in a variable? (when doing mkir, put it in an array of dirs to be removed at cleanup).

For me, you can close if the patch is tested and found working.

Anyway, thanks for investigating the issue.

BTW, when this happened, I was experimenting with a very simple setup: no build step, just copy files from the current directory. The solution was to set $build_root='.' A simple sample like that would be appreciated (no configure, no build, just copy files).

Regards // ATS

Hm... although i wrote erase.diff I don't think it really fixes this issue. IMO cleanup.diff is the better patch (explained in my previous comment).

Let's assume the user changes $build_root to /home/<user>/ and his project is in /home/<user>/<project>/. makepackage would then delete the whole home directory!

If a user sets the build_root himself, he knows what he's doing and should cleanup himself. Deleting anything else other then /tmp/apkg-* directories will always be a security risk i think.

The problem with cleanup.diff is that it removes all cleanup functionality, and we don't want that.

I have also been thinking along the lines of anything-not-in-$TMP shouldn't be deleted.

Maybe (pseudo code):

# Resolve any symlinks, relative directories and strip trailing/duplicate slashes etc.
$build_root = get_canonical_path($build_root)
# Whack it if it's inside $TMP
if $build_root.StartsWidth($TMP) && $build_root != $TMP then
   rm -rf $build_root
end if

I will cook up something this week unless you beat me to it Jan Niklas :)

But doesn't this line at the of CleanUp? do the job? rm -fr "${TMP}"/apkg-*

Replying to Jan-Nik:

But doesn't this line at the of CleanUp? do the job? rm -fr "${TMP}"/apkg-*

It does. Didn't notice that line was there.

Okay here's another patch, which checks if the directories are starting with ${TMP}/. Although I think cleanup.diff is okay, too.

I also changed CVS to SVN where is it mentioned in makepackage. (Completely different thing, sorry i was too lazy to create another patch)

Ok, safe.diff looks good - applied in r2447, Closing as fixed.