various performance optimizations #926
Conversation
|
Nice exploration, good to see someone dig in here. Sadly, I dont think its going to be this straightforward. The reason for multiple If we want to go to a single pass we need a way for plugins to invalidate and refetch some packages, this didnt exist in One of the ideas we threw around at the time was using one of the long running language servers, which should handle reloading packages better - eg gopls or guru. But we would need to be careful that |
| @@ -92,24 +90,39 @@ func ImportPathForDir(dir string) (res string) { | |||
|
|
|||
| var modregex = regexp.MustCompile("module (.*)\n") | |||
|
|
|||
| type NameForPackage struct { | |||
| cache *sync.Map | |||
There was a problem hiding this comment.
Why did this need to be moved out of a private global?
There was a problem hiding this comment.
Technically it didn't need to be moved, but it was weird to me to use global state when we were already creating a struct to hold the packages field.
In addition to the fact that global state is a dangerous anti-pattern, I suspect that this would also have caused issues if I wanted to use the same binary to generate multiple graphql servers.
This is not a PR that's meant to be merged. It's just a POC of some performance optimizations discussed in #918