command gets re-executed even with no change in input files

[SanderElias]

This might be me misunderstanding the nuances in the config. However, the command gets called every time I run npm run buildDeps

     "buildServer": {
      "command": "tsc --pretty --project server/tsconfig.server.json",
      "clean": "if-file-deleted",
      "files": [  "./server/**/*.mts", "./server/**/*.html",  "./server/tsconfig.server.json"  ],
      "output": [  "./dist/server" ]
    },
    "sharedSock": {
      "command": "ng build --configuration production sharedSock",
      "clean": true,
      "files": [ "./projects/shared-sock/src/lib/**/*.ts"  ],
      "output": [ "./dist/shared-sock"  ]
    },
    "buildDeps": {
      "dependencies": [  "buildServer", "sharedSock"  ]
    },

I tried quite a number of things, but the command for the "sharedSock" dependency gets executed every time.
It might be me misunderstanding the setup, but it should skip the command if there is no change in the files of the "sharedSock" right?
I looked at #238 which pointed me to #245, but I seem to already have taken those steps?

[aomarks]

Based on the config shown, it seems like you should be getting fresh skips/caching. Is there anything more to know about how you are running the command? If you directly run npm run sharedSock twice in a row, does it execute twice?

[SanderElias]

@aomarks Yes, it does execute twice.
In watch mode, it will execute on every change, including changes in the server

[aomarks]

(Sorry, I incorrectly referenced this issue which auto-closed it)

[milesrichardson]

I had a similar issue and was able to solve it by adding some exclude patterns to output. In my case I was generating .generated.ts files for any .ts file that includes some GQL snippet that graphql-codegen wants to use. The problem was that on the second run, wireit would report the output as "fresh", but copy files from the cache, and in doing so, delete the .generated.ts files.

The solution was to add two entries to output: !build, which was a simple oversight by me, but also weirdly !.wireit, which seemed like a bug. FWIW, this was in a child package, not the root.

Unfortunately my requirements meant I had to glob every file and remove the irrelevant entries. This was easy to do for input, since I could debug it by reading the .wireit/*/fingerprint file, which enumerates all input files resulting from evaluation of the files glob. But it does not enumerate all output files (instead, it enumerates the output glob patterns). This makes sense from a fingerprinting standpoint, but it made it difficult to debug an overzealous glob.

In general my frustration with tools like this is the opaque nature and slight idiosyncrasies of the various globbing libraries. It seems like every tool needs its own include pattern, and **/*.ts can mean something totally different in a config file for wireit or graphql-codegen.

I was able to debug my files patterns by reading the .wireit/*/fingerprint file, which enumerates the evaluated result of the files glob, but only lists the patterns for output. That makes sense from a fingerprinting standpoint, but it makes it difficult to debug an overzealous glob in output. (EDIT: I just noticed that these output files are enumerated in manifest – nice! FYI .wireit does not appear in that manifest file, but adding !.wireit to build reliably fixes my issue described above)

I think wireit could help make this less painful, by:

• Adding an environment variable to increase verbosity and log reason for invalidation of a dependency, kind of like how tsc --verbose will output "Project X is out of date because oldest output Y is older than newest input Z".
• Adding some kind of debug command or environment variable to print a list of evaluated glob expressions, so that we can check that our ordered array of regex definitions is selecting the expected set of files