[REQUEST] copy2fs

[jls]

Hi
I'm trying to compile a program, but I get error due to files being in different positions of /tmp/tcloop:
the .so files are in tcloop of the dev but the .so.1 are in the tcloop of the non dev.

[jls]

Solved importing in the same sce both dev and non-dev packages

[Jason W]

Normally I like to fix these kinds of things with startup scripts, but the option to copy2fs can be useful.  I will look at adding that option in. 

Which files are causing trouble?  I would like to fix them with startup scripts anyway.

[jls]

libelementary not being in the same sce of libelementary-dev

[Jason W]

I see with the libelementary-dev package.  This brings up a larger issue or solution in that perhaps all symlinks of an SCE should be copied to the filesystem rather than the links being linked, and their relative place under /tmp/tcloop then causing issue.  Would cost no more space and could prevent a lot of this kind of thing.  I will ponder that.

[Jason W]

Actual copy2fs would run great risk of locking up folks machines due to the nature of the size of the SCEs, would need a check and all like sce-import has.

But the problem I see is that with the "cp -ais" command used by sce-load, symlinks are linked and not copied as is.  And that is what is causing what you see and who knows how much else in other packages. 

I  am working on a way to copy symliks preserved as is to the filesystem while still able to just link all other files.  And without a performance cost.

[Jason W]

Ok, I have made some progress with the use of the find command combined with a tar pipe to copy symlinks preserved as they would be first before cp command would copy over as links all other files without overwriting the links that were already correctly copied over.  The find and tar pipe takes .75 seconds to copy over all 21,000+ symlinks in my 1.8GB main SCE.  That SCE takes a better part of a minute to load normally, more or less I have not timed it, so there is almost no performance penalty.

This should solve a LOT of errors like we have seen and prevent the use of starup scripts to have to copy over individual broken or wrong symlinks as we find them.  This method would be advantageous for Core as well, since symlinks should be copied as is to the live file system and not a symlink created pointing to a symlink buried in the mounted /tmp/tcloop area. 

Hopefully I will be able to wrap this up and upload the results tonight.

[Jason W]

Ok, fix applied.  Symlinks in packages get copied in their original state to the file system.  All x86 dCores uploaded to release_candidates, please test in the manner in which it was failing before with the -dev package in a separat SCE than the original.

[jls]

re-separeted dev and non dev and now it works!

[Jason W]

Great, that is what I was hoping to hear!