Ignore case in package directory

[vbraun]

The python tarball contains a upper-case Python-x.y.z directory. Our build system does not handle that case. Worse, it works differently on OSX and real OS'es because of filesystem case sensitivity.

Component: build

Author: Volker Braun

Branch/Commit: 9b31585

Reviewer: John Palmieri, Leif Leonhardy

Issue created by migration from https://trac.sagemath.org/ticket/16415

[vbraun]

Author: Volker Braun

[vbraun]

Branch: u/vbraun/ignore_case_in_package_directory

[vbraun]

Commit: e9aa3c6

[vbraun]

New commits:

e9aa3c6

ignore case when renaming source directory

[nexttime]

comment:4

Is this supposed to be a temporary (=ad-hoc) solution?

In the long run, the actual folder name should probably be part of Sage's "package metadata".

[vbraun]

comment:5

If there is a single top-level directory then we don't need to store the name (and manually maintain) it elsewhere. There should be some checking, of course. And if you don't have a single top-level directory then I'd consider that to be an upstream bug.

[nexttime]

comment:6

Replying to @vbraun:

If there is a single top-level directory then we don't need to store the name

Yep, but that's NYI.

(and manually maintain) it elsewhere.

Well, we "manually" maintain the "checksums" for each specific upstream tarball anyway (in a central repo, tied to specific Sage versions, which is IMHO a bad thing to do anyway).

There should be some checking, of course.

Sure.

And if you don't have a single top-level directory then I'd consider that to be an upstream bug.

:-)

[nexttime]

comment:7

P.S.: FWIW, requiring foo-x.y.pN.spkg to have a foo-x.y.pN/ folder was used as a consistency check (although I didn't like it, as it just made testing spkgs slightly harder).

[sagetrac-git]

Branch pushed to git repo; I updated commit sha1. New commits:

9b31585

bash 4.1 workaround

[sagetrac-git]

Changed commit from e9aa3c6 to 9b31585

[vbraun]

comment:9

Also removed the hiding-of-errors, I want to know which packages don't work...

[nexttime]

comment:10

Probably immediately go with find . -type d -a -not -name . -exec mv {} src \; 2>/dev/null ;-) (and afterwards test -d ...)

(Haven't checked right now, but I guess -iname isn't portable.)

[vbraun]

comment:11

Besides not portable, its IMHO not readable. The whole contraption should be rewritten in Python with some proper doctesting.

[nexttime]

comment:12

Replying to @vbraun:

Besides not portable, its IMHO not readable.

Well, the find version above is portable, although I'd just use it to get the folder name. (Or some flavour of (cd * && basename `pwd`).)

The whole contraption should be rewritten in Python with some proper doctesting.

In Lisp, please.

Anyway, I'd really explicitly check whether src/ afterwards exists (and is a directory); we already have

    echo "Finished extraction"

    cd "$PKG_NAME"
    if [ $? -ne 0 ]; then
        echo >&2 "Error: after extracting, the directory $PKG_NAME does not exist"
        exit 1
    fi

right below for legacy spkgs.

Just out of curiosity: Which bash version(s) did the original change work with?

I tried older (3.2.) as well as newer (4.3.11) bashs; with none of them it worked.

[vbraun]

comment:13

$ bash --version
GNU bash, version 4.2.47(1)-release (x86_64-redhat-linux-gnu)

Replying to @nexttime:

In Lisp, please.

Sage is a Python project. And Python is actually pretty good for sysadmin-type scripts. Unlike lisp. And unlike Bash it can be tested in a sane way.

[nexttime]

comment:14

Replying to @vbraun:

$ bash --version
GNU bash, version 4.2.47(1)-release (x86_64-redhat-linux-gnu)

FWIW, "${PKG_NAME_UPSTREAM%.tar*}" doesn't work with

GNU bash, version 4.2.28(1)-release (x86_64-redhat-linux-gnu)
Copyright (C) 2011 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>

either. (So it's probably a bug in your bash that it works for you. ;-) )

[nexttime]

comment:15

I'd still add a check (with an appropriate error message) that src/ afterwards really exists (and is a directory ;-) ); cf. my comment above.

[vbraun]

comment:16

We don't need a double check, either building the package will break if the rename fails or the rename is not necessary (and, really, wtf do we rename the tarball directory to start with, afaik nobody else does that). In particular, there is no need to check that the renamed directory is still a directory.

[jhpalmieri]

comment:17

Looks okay to me.

[jhpalmieri]

Reviewer: John Palmieri, Leif Leonhardy

[vbraun]

Changed branch from u/vbraun/ignore_case_in_package_directory to 9b31585