Code Quality: Improve CommandManager

[KrisVandermotten]

Resolved / Related Issues

• Closes Code Quality: Inefficiencies in CommandManager #18136

Steps used to test these changes

1. Inspect the generated files, and compare them to the previously generated files
2. Run Files and use commands

[yair100]

Thank you for the PR.

[0x5bfa]

LGTM. This looks to be incremental now.

[KrisVandermotten]

This looks to be incremental now.

It is indeed.

Do you want me to update the StringBuilder logic in StringsPropertyGenerator using raw strings too (in a separate PR, obviously)?

[yair100]

Does this have a positive impact on startup performance?

[0x5bfa]

Generators run on VS threads at compilation time, no.

Do you want me to update the StringBuilder logic in StringsPropertyGenerator using raw strings too (in a separate PR, obviously)?

I think it's already better.

[KrisVandermotten]

I think it's already better.

I think it's indeed improved in this PR. But my question was about #18161, where I did not touch the StringBuilder code (yet). Do you want me to make a new PR, that improves that code similarly to the improvements here?

[KrisVandermotten]

Does this have a positive impact on startup performance?

Not sure what you are referring to?

This PR does have a positive impact on startup performance. The newly generated code is more efficient than the previously generated code. Just to be clear, I have no plans for further changes to this PR.

On the other hand, changing the way in which the code is generated, i.e. replacing multiple calls to AppendLine by a single call to Append with a multiline raw string in the generator, has no impact on the generated code whatsoever. This is primarily about readability of the generating code. The generation is also slightly more efficient, but the generated code is the same. I already did this in this PR, but not (yet) in #18161.

[yair100]

Thanks for confirming. Did you run any benchmarks to compare the difference in startup performance? If it's noticeable, I can include a mention in the next update changelog.

[KrisVandermotten]

I did not run any formal benchmark. Measuring cold startup time correctly is not exactly trivial.

I would not expect the speedup to be immediately noticeable with the naked eye. There are a lot of things going on during startup, and building the CommandManager is just one of them. You know, Amdahl's law.

[yair100]

Thanks again.