Intended install location for python scripts

Hi all,

I started digging into the ‘make install’ python script installation issue listed on another thread. Here’s the ticket for it:

http://llvm.org/bugs/show_bug.cgi?id=18124

I’ve added some comments to the bug. My initial digging leads to a few questions:

  1. Where do we expect a configure/make/make install to place the python scripts? I was under the impression the scripts should be installed in the same directory tree implied by the configure --prefix arg, which is also what a lldb -P reports. However, what I’m finding is that the ‘make install’ step is trying to install the python scripts into the python interpreter’s dist-packages location. I’m getting these failures on install because I’m not doing a ‘sudo’ on my make install.

llvm[4]: Installing Debug+Asserts LLDB python modules
mkdir: cannot create directory /usr/lib/python2.7/dist-packages/lldb': Permission denied Error: Unable to find or create /usr/lib/python2.7/dist-packages/lldb make[4]: Leaving directory /usr/local/google/home/tfiala/llvm/work/build-debug/tools/lldb/source/Interpreter’

Do we want the python scripts installing into the python executable’s dist-packages or the location reported by lldb -P? My first stab would be it should go in the installation prefix paths as I would not expect any code to be installed outside of the prefix and, possibly, require a sudo when I’m trying to do everything locally without elevated privileges.

  1. Should ‘make install’ fail if we fail to copy over python files? My first take is that a failure to install the scripts should qualify as an all-out failure and cause make to fail (i.e. fail early, fail loudly); otherwise, we can get quite a bit past the root cause of scripting-related failures without realizing what’s broken.

Any opinions on this? My straw-man answers are:

  1. Install into the tree implied by configure --prefix, not into the python dist-packages dir. (A close second would be to specify a new configure path variable so the user can configure, possibly adjusting the output of lldb -P to match).

  2. Change makefiles so that failure to copy scripts fails the install.

I’ll fix up whatever we decide here if it requires a change.

Thanks!

Sincerely,
Todd Fiala

Hi all,

I started digging into the 'make install' python script installation issue listed on another thread. Here's the ticket for it:

http://llvm.org/bugs/show_bug.cgi?id=18124

I've added some comments to the bug. My initial digging leads to a few questions:

1. Where do we expect a configure/make/make install to place the python scripts? I was under the impression the scripts should be installed in the same directory tree implied by the configure --prefix arg, which is also what a `lldb -P` reports. However, what I'm finding is that the 'make install' step is trying to install the python scripts into the python interpreter's dist-packages location. I'm getting these failures on install because I'm not doing a 'sudo' on my make install.

    llvm[4]: Installing Debug+Asserts LLDB python modules
    mkdir: cannot create directory `/usr/lib/python2.7/dist-packages/lldb': Permission denied
    Error: Unable to find or create /usr/lib/python2.7/dist-packages/lldb
    make[4]: Leaving directory `/usr/local/google/home/tfiala/llvm/work/build-debug/tools/lldb/source/Interpreter'

Do we want the python scripts installing into the python executable's dist-packages or the location reported by lldb -P?

Yes, this would allow multiple installs of LLDB to co-exist without stomping on the previous LLDB's python modules (which could be very different).

My first stab would be it should go in the installation prefix paths as I would not expect any code to be installed outside of the prefix and, possibly, require a sudo when I'm trying to do everything locally without elevated privileges.

2. Should 'make install' fail if we fail to copy over python files? My first take is that a failure to install the scripts should qualify as an all-out failure and cause make to fail (i.e. fail early, fail loudly); otherwise, we can get quite a bit past the root cause of scripting-related failures without realizing what's broken.

Any opinions on this? My straw-man answers are:

1. Install into the tree implied by configure --prefix, not into the python dist-packages dir. (A close second would be to specify a new configure path variable so the user can configure, possibly adjusting the output of lldb -P to match).

The --prefix path sounds right to be for coexistence.

2. Change makefiles so that failure to copy scripts fails the install.

Agreed.

Excellent. Thanks, Greg!

I’ll get a patch together following that approach unless anything further develops here.