Faster __vsc_escape_value for bash

[hsfzxjy]

Fix #190255

This PR presents two faster escaping functions as the replacement of the original slower one.

The first one escapes each byte 0xab into '\xab', which is most scalable and has promising runtime
efficiency, except that it relies on external commands od and tr.

The second one utilizes bash builtin string substitution, which is much faster and has zero dependency, except that it escapes only '\\' -> '\\\\' and ';' -> '\x3b', which scales up badly for more patterns.

Currently it defaults to use the second function, but if env VSCODE_SHELL_FULL_ESCAPE is set and od, tr are
available, it will switch to the first one.

We may discuss more on the choice of the implementation.

[Tyriar]

Is VSCODE_SHELL_FULL_ESCAPE required because you're concerned about breakages?

[hsfzxjy]

I think both solutions have their pros and cons, so I temporarily make one of them as default and the other opt-in controlled by a newly introduced flag VSCODE_SHELL_FULL_ESCAPE, but this is apparently not the most optimal solution. I am open to consider any further advice as you see fit.

[Tyriar]

Should we perhaps select a magic number for $1 length and opt for the new approach when it's above that threshold?

[hsfzxjy]

Sounds reasonable. What about make it 2000 bytes? This is a point where I can observe around 0.1s delay with the old method.

[Tyriar]

On your machine how long does the new method take? 100ms is quite large still

[hsfzxjy]

Some benchmarks

Input size (bytes)	string sub(new default)	od+tr(new opt-in)	old
2k	1ms	5ms	107ms
100k	33ms	48ms	>1 min

I think with the new method the overhead is tolerable even with quite large input.

[Tyriar]

They both seem very fast. If I understand right, the od+tr method is superior as it escapes more correctly. Since perf is comparable, I think we should default to the od+tr method if they're available without an opt-in variable, then fallback to the more naive method when they're not available.

[hsfzxjy]

Sure. I will change accordingly. Shall we still set up the magic number?

[Tyriar]

Sure, let's make it 2k

[hsfzxjy]

BTW, is /dev/null always available for the execution context of shellIntegration-bash.sh? The script is using command -v od &>/dev/null to detect the availability of od.

Edit: never mind. I found solution to get around it.

[hsfzxjy]

The latest push contains these further modification:

1. od+tr becomes the default solution for faster escaping, and if od or tr is not available, we fallback to use the string substitution solution.
2. The old __vsc_escape_value function is still used when the length of $1 is below 2000. Otherwise we switch to the new faster function.

[Tyriar]

Thanks @hsfzxjy, this looks good. I'll wait until next month to merge so we get some testing in Insiders before it goes out.

[hsfzxjy]

Accidental force push dimissed the reviews. Sorry! 🙏

[Tyriar]

This ended up getting reverted due to a hang #190255 (comment)