fix: support git+ IRIs

[andrius-puksta-sensmetry]

• Fix support for git+ URLs.
• Make git clones shallow

Tested sysand clone with the following, all pointing to a git repo:

• git+file
• git+https for public repo (https://github.com/andrius-puksta-sensmetry/test-sysml-project)
• git+ssh for the same repo, both public and private
• ssh for the same public repo
• https for the same public repo

Notes by Erik

Here is how I ran the same tests manually. I ran into unrelated networking issues that confused me initially though.

# git+file test OK
d=$(mktemp -d) && cd "$d" && cargo run --manifest-path ~/dev/sensmetry/sysand/Cargo.toml clone git+file:///home/erik/dev/consideratio/sysand-test-project

# git+https test OK
d=$(mktemp -d) && cd "$d" && cargo run --manifest-path ~/dev/sensmetry/sysand/Cargo.toml clone git+https://github.com/consideratio/sysand-test-project

# git+ssh test
# - several users likely mistakenly use `:` between host and path, and get "failed to canonicalize iri", an error message that could be improved
d=$(mktemp -d) && cd "$d" && cargo run --manifest-path ~/dev/sensmetry/sysand/Cargo.toml clone git+ssh://git@github.com/consideratio/sysand-test-project

# https test OK
d=$(mktemp -d) && cd "$d" && cargo run --manifest-path ~/dev/sensmetry/sysand/Cargo.toml clone https://github.com/consideratio/sysand-test-project

# ssh test
d=$(mktemp -d) && cd "$d" && cargo run --manifest-path ~/dev/sensmetry/sysand/Cargo.toml clone ssh://git@github.com/consideratio/sysand-test-project

[consideRatio]

I see ssh://git, but there seems to be more things supported not examplified here, such as http/https/file sections

[andrius-puksta-sensmetry]

That can wait. ssh://git was specifically mentioned here to prevent people from trying to use the shorthand, as the shorthand is directly given by the GH UI and is the simplest way to refer to a repo (no messing with authentication vars, sysand uses standard OS ssh keys).

I would like to add support for shorthand ssh syntax at some point, as it's the standard way of referring to git repos.

Many more examples could be added here, but I'll leave it for the future:

• directly referring to a file/directory in a GitHub/GitLab repo
• public and private repos, explaining authentication (ssh/bearer token)
• multiple ways of referring to the same stuff: as a git repo/as a file or folder in the repo/as a release asset of the repo/api vs non-api (at least for GitHub)

[consideRatio]

I'm trying to figure out what is supported, its not obvious to me in order to review and test!

git+file
git+ssh for the same repo, both public and private
ssh for the same public repo
https for the same public repo

# git+file test OK
mkdir -p /tmp/a && cd /tmp/a
cargo run --manifest-path ~/dev/sensmetry/sysand/Cargo.toml clone git+https://github.com/consideRatio/sysand-test-project

# git+file test OK
mkdir -p /tmp/b && cd /tmp/b
cargo run --manifest-path ~/dev/sensmetry/sysand/Cargo.toml clone git+file:///home/erik/dev/consideRatio/sysand-test-project

# git+ssh test
# - several users likely mistakenly use `:` between host and path, and get "failed to canonicalize iri", an error message that could be improved
# - case sensitive in path, can get stuck if redirected (consideratio instead of consideRatio)
mkdir -p /tmp/c && cd /tmp/c
cargo run --manifest-path ~/dev/sensmetry/sysand/Cargo.toml clone git+ssh://git@github.com/consideRatio/sysand-test-project

Ran out of time to work this further today, gtg from work now!

[andrius-puksta-sensmetry]

I found that git+file:// was case sensitive

The scheme string itself, or the path?

[consideRatio]

I found that git+file:// was case sensitive

The scheme string itself, or the path?

The path -- i had written consideRatio but my folder was named consideratio

[andrius-puksta-sensmetry]

# lowercase R in my github username - gets stuck indefinitively
# if i do git clone, this isn't a problem however

Look into this. We use gix, which should match git's behaviour. If fixing requires changes to gix, should at least try to error out and not get stuck.

[andrius-puksta-sensmetry]

I found that git+file:// was case sensitive

The scheme string itself, or the path?

The path -- i had written consideRatio but my folder was named consideratio

Why is case sensitivity unexpected? This is a path, and so follows rules of the filesystem/OS.

[consideRatio]

Why is case sensitivity unexpected? This is a path, and so follows rules of the filesystem/OS.

Ah, got thinking it wasn't case sensitive on my system, but it was.

[consideRatio]

I had intermittent network issues in the office influencing SSH greatly, but it was unrelated.

I also conclude that you can't do a trivial ":" check to help users towards using / as a separator between host and path, as you may use : like host[:port]/pat, so skipping it for now, and skipping adding tests for now also.