replace bash subcommands with separate shell spawns

[rvagg]

As per discussion in #66

I need to think about this a bit more, there's likely going to be overhead for non-windows users because this replaces all Bash subcommands with separate spawn(shell) executions and piecing the output together to form the final git log command.

It also leaves this repo in a 1/2 promisified state which is a bit awkward.

But, let's see what CI does.

Closes #63, fixes #54

[rvagg]

👍 passing on windows, though I've removed 8.x in the latest commit, stream.pipeline() is a 10.x feature and util.promisify() is improved in 10.x too.

[XhmikosR]

Remember that there's a case where things worked on Windows CI but still failed on vanilla Windows cmd :)

Please ping me when you think this is ready so that I try it on my Windows VM.

[XhmikosR]

@rvagg just tested this on my Windows VM, with /origin/xmr-pack-gitexec-up merged in this branch and seems to work fine.

C:\Users\xmr\Desktop\changelog-maker>ver

Microsoft Windows [Version 10.0.18362.418]

C:\Users\xmr\Desktop\changelog-maker>node -v && npm -v
v12.12.0
6.11.3

C:\Users\xmr\Desktop\changelog-maker>npm t

> [email protected] test C:\Users\xmr\Desktop\changelog-maker
> npm run lint && npm run unit


> [email protected] lint C:\Users\xmr\Desktop\changelog-maker
> standard


> [email protected] unit C:\Users\xmr\Desktop\changelog-maker
> tape test.js

TAP version 13
# test basic commit block
ok 1 should be equal
# test filter-release
ok 2 should be equal
# test simple
ok 3 should be equal
# test group, semver labels, PR-URL
ok 4 should be equal
# test simple group, semver labels, PR-URL
ok 5 should be equal
# test blank commit-url
ok 6 should be equal
# test blank commit-url
ok 7 should be equal

1..7
# tests 7
# pass  7

# ok

BTW, so that we are safe maybe we should close #66 and you rebase this branch to include the patch instead?

EDIT: NVM you updated to 2.0.1 here too.

[rvagg]

I can measure a very small performance impact by this. In the order of 10-20ms, and it's going to depend a bit on I/O speed.

I think that having this work on Windows is probably worth the price. Unwinding some of the complexity of the git commands is also going to be a bonus here I think--the way it's currently structured makes it quite unfriendly for contributors I think. I would like some other opinions on that though.

This change will be semver-major because we're dropping support for older Node. I wouldn't mind getting at least a new ghauth() in before a release too, that should mostly take care of the TODO in here but will be a separate PR because it needs a refactor of collect-commit-labels.js.

[XhmikosR]

Yup, I noticed a small perf impact too. But I agree with your points.