Looks okay if you've tested it. But as with #137, wouldn't basename be a cleaner way of accomplishing the last(split(...)) ?

Switched to basename(). I also verified code as of 98878bd still downloads and extracts files without errors on my .7 and .6 installs.

This approach also handles filenames containing + / %2B, correct?

I don't handle the filenames any differently than the code that was in place does. The only change that has been made is that any directory parts of the filename are now chopped off I.E. if the file is named noarch/happy_package.1.1.1.rpm I trim that to just be happy_package.1.1.1.rpm. So if a file was named noarch/package+%2B-1.1.1.rpm I would change it to package+%2B-1.1.1.rpm.

The old code was in essence (combining two lines into one):
cpio = splitext(joinpath(cache,escape(path)))[1]*".cpio"
My code is:
cpio = splitext(joinpath(cache, escape(basename(path))))[1] * ".cpio"

Thanks. If you've tested it and it works, looks good to me. @tkelman Fine with you too?

This does look like a duplicate of #137 but is perhaps better than what I had done, I don't know. I ran this and it worked for me on v0.7 (combined with #136).

Unlike #137, this looks like it should avoid breaking anything on 0.6

Meaning, OK to merge? :-)

bump hdf5 build fails without this patch

would also be good to tag after this

@tkelman Bump? I'm afraid you're the only one which knows this package enough. :-)

Should really set up appveyor here to be sure but this is likely safe in its current form.

Agreed. I did test locally that this pr worked. BTW i did try to get appveyor, but didn't add a proper cleaning step. I could look into that again, but no guarantees on when. For reference #95

OK, merging since anyway this doesn't change the behavior on 0.6, and it can't be worse on 0.7.

I still have the issue also mentioned by @amellnik
When installing Cbc I run into the following issue:
the rpm file is called

noarch%2Fmingw64-libstdc%2B%2B6-7.2.0-3.2.noarch.rpm
The filename got escaped, due to this statement
path2 = joinpath(cache,escape(path))

However the file in that archive is called

mingw64-libstdc++6-7.2.0-3.2.noarch.rpm
When determing the cpio filename, it is assume this is the same file name, which is not correct as it does not take the escaping into account, indeed for .7 we have
cpio = splitext(joinpath(cache, escape(basename(path))))[1] * ".cpio"
this should be replaced with
cpio = splitext(joinpath(cache, basename(path)))[1] * ".cpio"
In my local version this does the trick.

@nalimilan @tkelman Can you have a look at my comment above, this kinda stops my 0.7/1.0 transition ?

Dunno. What changed?

in the download from the filename gets escaped. Then the compressed file is unpacked and it looks for .cpio. however in the compressed file the filename has not been escaped.
I'll give an example, during the building of Cbc http://download.opensuse.org/repositories/windows:/mingw:/win64/openSUSE_13.2/noarch/mingw64-libstdc++6-7.2.0-3.2.noarch.rpm
gets downloaded to a local file called
noarch%2Fmingw64-libstdc%2B%2B6-7.2.0-3.2.noarch
in the cache folder
the 7zip then is called to unpack file mingw64-libstdc%2B%2B6-7.2.0-3.2.noarch.cpio
but that file does not exists, it is called mingw64-libstdc++6-7.2.0-3.2.noarch.cpio

13.2 is very old and shouldn't be referenced any more... the escaping must've been added for a reason and presumably worked and was necessary before?

My bad, I just grabbed a random SUSE version to make my case, CBC uses 42.2 actually.
I do not know exactly why the downloaded file needs the escaping, I know windows does not need it, not versed enough in other OS's to be say why.
The behavior of 7z seems to have changed in 0.7/1.0. In 0.6 using the wrong filename for extraction did still work, it does no longer in 0.7/1.0. I have read a reference to this somewhere, will dig it up and post when I found it.

There was a change in 7z version and behavior between julia 0.6 and julia 0.7, but that happened before February/March when this PR was worked on. I'm under the assumption the author and other commenters on this PR had tested it on 0.7-DEV at that point. I'm asking what changed since February.

I tested this with a .6 and .7 later than 2181, when the 7 zip version changed. The behavior described exactly matches what this PR fixes. Are you accidentally using a cached version of WinRPM prior to when this merge happened?

I have cleared everything, problem persists

As you can see , the cpio file is extracted correctly from the RPM, however the name is not escaped (note the ++)
And here is where things go wrong (line 467)

info("Extracting: ", name)
cpio = splitext(joinpath(cache, escape(basename(path))))[1] * ".cpio"

The file name cpio is built using an escaped name, the unzipped cpio file is unescaped, this then leads to the extract command $exe7z x -y $cpio -o$installdir failing

I see what you are saying. The patch fixed all the other cases because the filenames did not have any special characters in them. However, your c++ filename gets mangled because it has html escape characters in it.

I agree the solution should be to just remove the escape() however I fear without testing this might break something else as this code is very arcane. My .7 Julia install is a bit out of date but I will see if I can get a new PR in with your proposed changes.

I am willing to do the PR.
One approach would be to test if the file exists (with and without escaping). It's not elegant, but for arcane code it might be an acceptable (=safe) approach.

Sounds like a good approach that will ensure the file gets handled appropriately. I'll lave it to you to do the PR then.

OK, I'll tackle that. Not that much experience on git, so it's a good opportunity to learn something.

Made the PR #157