Skip to content

Latest commit

 

History

History
336 lines (232 loc) · 9.49 KB

README.md

File metadata and controls

336 lines (232 loc) · 9.49 KB
Raw

leopard.sh and tiger.sh

Package managers for PowerPC Macs running OS X Leopard (10.5) and Tiger (10.4), written in Bash 😱

In a nutshell:

$ curl -O http://leopard.sh/leopard.sh
$ chmod +x leopard.sh
$ ./leopard.sh quake.app-1.1
$ open "/Applications/GLQuake 1.1"

or:

$ curl -O http://leopard.sh/tiger.sh
$ chmod +x tiger.sh
$ ./tiger.sh quake.app-1.1
$ open "/Applications/GLQuake 1.1"

Binary package support:

I build pre-compiled binary packages optimized for the following platforms:

G3 G4 G4e G5/32 G5/64
Leopard - - ✅ ✅ ✅
Tiger ✅ ✅ ✅ ✅ ✅

The above processors can be understood as the following gcc flags:

  • G3: -mcpu=750
  • G4: -mcpu=7400
  • G4e: -mcpu=7450
  • G5/32: -mcpu=970 -m32
  • G5/64: -mcpu=970 -m64

Optimization flags used:

G3 G4 G4e G5/32 G5/64
Leopard -O2 -O2 -O2
Tiger -Os -Os -O2 -O2 -O2

Installation:

On your Mac, open up Terminal.app.

Download the leopard.sh Bash script:

$ curl -O http://leopard.sh/leopard.sh

or, for Tiger users:

$ curl -O http://leopard.sh/tiger.sh

Now make it executable:

$ chmod +x leopard.sh

Now run it:

$ ./leopard.sh

During the first run, some setup will be performed, e.g. creating /opt, downloading a few dependencies, etc. You'll be prompted for your password, because a few of these commands need root access.

Also, you'll be prompted to visit https://leopard.sh/md5 to verify that the script wasn't tampered with during download. Because the version of Safari which shipped with Leopard and Tiger doesn't support modern SSL, this step has to be performed on either a modern PC or on your smartphone.

List the available packages:

Fetch the list of the available packages:

$ ./leopard.sh
Available packages:
adium.app-1.3.10
clisp-2.39.20210628
coreutils-9.0
gcc-4.9.4
gettext-0.20
gmp-4.3.2
gzip-1.11
handbrake.app-0.9.1
...

These names, e.g. gzip-1.11 are called "pkgspecs".

You'll also notice that there are two kinds of pkgspecs supported:

  • Unix software, e.g. gzip-1.11
  • Mac applications, e.g. adium.app-1.3.10

Note that the pkgspecs are printed to stdout, while the "Available packages:" header is printed to stderr. This means we can pipe leopard.sh into grep:

$ leopard.sh | grep gzip
Available packages:
gzip-1.11

We could also use grep to list just the Mac applications:

$ leopard.sh | grep .app
Available packages:
adium.app-1.3.10
chicken.app-2.2b2
clozure-cl.app-1.4
handbrake.app-0.9.1
interwebppc.app-rr1
quake.app-1.1
textwrangler.app-3.1
vlc.app-0.9.10
xbench.app-1.3
...

The Unix geeks will note that technically we should grep for '\.app', not .app.

Install a package:

To install a package, call leopard.sh with a pkgspec:

$ leopard.sh gzip-1.11

Note: currently, you have to use the full pkgspec, i.e. you can't do something like leopard.sh gzip. I'll implement some sort of pkgspec aliases soon, but in the mean time you could use a subshell as a work-around:

$ leopard.sh $(leopard.sh | grep gzip)

The above command will fetch and run http://leopard.sh/leopardsh/scripts/install-gzip-1.11.sh.

That script will look for a pre-compiled binary package (a "binpkg") which matches your CPU type, e.g. a Mac with a G5 would try to fetch https://leopard.sh/leopardsh/binpkgs/gzip-1.11.leopard.g5.tar.gz and unpack it into /opt/gzip-1.11, and create symlinks in /usr/local/bin.

If for some reason a binpkg is unavailable, the script will instead attempt to compile gzip-1.11 from source. (However, I currently build binpkgs for every supported OS / CPU combo, so I'm the only one who will ever see this step).

Some pkgspecs depend on other pkgspecs, i.e. https://leopard.sh/leopardsh/scripts/install-gcc-4.9.4.sh

In that case, the script will call leopard.sh to install each of its dependencies, and so on, recursively.

Remove a package:

To remove a package, first clean up its symlinks from /usr/local/bin:

$ leopard.sh --unlink gzip-1.11

and then delete it from /opt:

$ rm -rf /opt/gzip-1.11

That's right: you remove packages from /opt manually. Note that leopard.sh does not maintain any sort of database or state tracking about what's currently installed (it simply checks if /opt/foo-1.0 exists), so in general you are free to muck with /opt in any way you see fit. That is, leopard.sh does not own /opt, it is a citizen of /opt.

If you had previously unlinked a package but hadn't actually deleted it, and now you'd like to re-link it, just run:

$ leopard.sh --link gzip-1.11

Project goals:

  • Performance

It should be snappier than homebrew, even on a slow G3.

  • Full binary package availability.

Every pkgspec should have a pre-compiled binary package available for every OS / CPU combo. Users should never wait on compilation.

  • Simplicity / hackability / conceptual flatness.

The problem with most package managers is that they have a very steep learning curve for contributing / hacking.

leopard.sh and tiger.sh will take the opposite approach: the installation of a package shall be encompassed entirely by two Bash scripts:

  • the orchestrator script: leopard.sh or tiger.sh
  • the package-specific script: e.g. install-gzip-1.11.sh

leopard.sh and tiger.sh are spritual brethren to the Linux From Scratch project.

Would you like to hack on some of the installer scripts yourself? It is as simple as:

$ cd
$ curl -L https://github.com/cellularmitosis/leopard.sh/archive/refs/heads/main.tar.gz | gunzip | tar x
$ export LEOPARDSH_MIRROR=file://$HOME/leopard.sh-main
$ # or, for tiger:
$ export TIGERSH_MIRROR=file://$HOME/leopard.sh-main

Bam, you're now running entirely from your self-contained copy of the project.

In fact, as long as you drop any needed source tarballs into ~/Downloads before running leopard.sh, you're now also completely offline-capable.

Project non-goals:

  • Supporting Intel Macs 😜

intel

Other options:

If leopard.sh and tiger.sh aren't for you, checkout these other package managers:

Mirroring

The http://leopard.sh server now runs rsyncd!

List the available directories via:

rsync leopard.sh::

The only exposed directory is html (as in /var/www/html), so you can create a mirror via:

rsync -av leopard.sh::html .

But in practice this would look something more like:

rsync -av --delete leopard.sh::html /var/www/html/mirrors/leopard.sh/

Support for mirrors which force HTTPS

Where appropriate, leopard.sh will use plain-HTTP URLs for a small performance boost (or in the case of a slow G3, not so small!).

However, some mirrors automatically redirect all plain-HTTP traffic to HTTPS. This defeats the point of using plain-HTTP, as you not only suffer the performance hit of using HTTPS, but on top of that an additional redirect from HTTP to HTTPS. It would have been better to use HTTPS to begin with!

In this case, you can export LEOPARDSH_MIRROR_NO_HTTP=1 (or export TIGERSH_MIRROR_NO_HTTP=1) to inform leopard.sh that it should not use plain-HTTP URLs.

Note that if you set LEOPARDSH_MIRROR_NO_HTTP and use an HTTPS-only mirror when intially installing leopard.sh, it will be unable to download download its dependencies, as /usr/bin/curl does not support modern HTTPS.

In this case, you can download tigersh-deps-0.1.tiger.g3.tar.gz using another computer, then place it in /tmp and run leopard.sh again.

Support for mirrors which don't allow HEAD

The HTTP HEAD operation can be used to fetch only the HTTP headers without downloading the contents of the URL.

leopard.sh uses HEAD to fetch the Content-Length of files before downloading them, and hands that size to pv to make its progress bar more informative.

However, some mirrors do not support the HEAD operation, causing spurious failures.

In this case, you can export LEOPARDSH_MIRROR_NO_HEAD=1 (or export TIGERSH_MIRROR_NO_HEAD=1) to inform leopard.sh that it should not attempt to fetch the size of files before downloading them.

Useful links for OS X PowerPC users:

Vintage Mac content: