terraform_deprecated_index: improve perf for files with many expressions

[bendrucker]

A coworker at @observeinc alerted me to a module where new versions of TFLint were running extremely slowly. Some initial investigation revealed that terraform_deprecated_index was spending a really long time when there were a very large number of expressions to walk in a module. In this case, there were a few large JSON objects being encoded with jsonencode and all the key/value pairs are expressions, though mostly static.

# times out
tflint --only terraform_deprecated_index

Most of that time was going into repeatedly calling runner.GetFile(), which in the example case was being called hundreds of times for the same file. There's no parser instance stored on the runner so each GetFile call meant the file was read from disk and parsed as HCL.

Since WalkExpressions calls GetFiles() anyway, calling it once before walking is a straightforward fix that brings the performance to an acceptable point. In the module in question, previously TFLint ran for multiple minutes on my Mac Studio at 200% CPU before I'd give up after 5m and interrupt it before it printed any output. After, it finishes in ~8s. Caching files after load would probably further boost performance by avoiding multiple rounds of loading from disk and parsing. The memory penalty would probably be worthwhile for the reduced CPU usage.

[wata727]

Makes sense. Thanks!

[wata727]

There's no parser instance stored on the runner so each GetFile call meant the file was read from disk and parsed as HCL.

In my understanding, the host process keeps the map of files in the memory, so I think there is no I/O or parsing cost.
https://github.com/terraform-linters/tflint/blob/v0.43.0/plugin/server.go#L64-L67

However, the RPC cost will not be negligible. This fix is perfectly valid.