Missing --dump-input flags

[AntonLydike]

Currently responsible for 14 / 69 remaining test failures:

• mlir/test/Conversion/AsyncToLLVM/convert-runtime-to-llvm.mlir
• mlir/test/mlir-cpu-runner/async-error.mlir
• mlir/test/mlir-cpu-runner/async-func.mlir
• mlir/test/Dialect/Async/coro.mlir
• mlir/test/Dialect/Async/async-to-async-runtime.mlir
• mlir/test/mlir-cpu-runner/async-value.mlir
• mlir/test/Dialect/Async/async-parallel-for-async-dispatch.mlir
• mlir/test/Dialect/Async/runtime.mlir
• mlir/test/Dialect/ControlFlow/canonicalize.mlir
• mlir/test/Dialect/Math/algebraic-simplification.mlir
• mlir/test/Dialect/Async/async-parallel-for-num-worker-threads.mlir
• mlir/test/Dialect/Async/async-parallel-for-seq-dispatch.mlir
• mlir/test/Dialect/SCF/one-shot-bufferize-allow-return-allocs-no-deallocs.mlir
• mlir/test/Dialect/Shape/remove-shape-constraints.mlir

[stanislaw]

The StrictDoc project now has all tests failing because of this, see strictdoc-project/strictdoc#1921 (comment).

Integration tests fail because the recently released filecheck 1.0.0 doesn't support --dump-input. This is a regression unrelated to this PR.

[AntonLydike]

Thanks for the pointer! I will look into this tomorrow, I have an exam later today!

[AntonLydike]

Sorry I didn't get to it yet, there's currently a lot going on and I'm missing stuff on my to-do list. I'll try to get this done this week!

[AntonLydike]

@stanislaw Can you tell me more how you use --dump-input in strictdoc? It seems that the flag is strictly used for debugging test failures, and the checking behaviour is unchanged by this flag.

Error messages in one field where I think diverging from LLVM filecheck is acceptable, e.g. printing variable bindings, CHECK-DAG regions, etc. (as long as it helps debugging matching issues).

[stanislaw]

Many of the StrictDoc tests use | filecheck %s --dump-input=fail.

This option doesn't change anything in how the checks are made. When it is activated, the full input provided/piped to filecheck is printed. This has been very useful for implementing the itests and debugging them in case of regressions.

P.S. I am still waiting for you to come back regarding the license topic. Please do not take too much time to consider it.

[AntonLydike]

Many of the StrictDoc tests use | filecheck %s --dump-input=fail.

This option doesn't change anything in how the checks are made. When it is activated, the full input provided/piped to filecheck is printed. This has been very useful for implementing the itests and debugging them in case of regressions.

Interesting! fail seems to be the default value for --dump-input, I'll think about the best way to proceed... I am unsure if the other options never (don't print error messages), help (kind of like --help, but for the output produced) and always (print the input to stderr if successful) are particularly useful modes, to be honest.

P.S. I am still waiting for you to come back regarding the license topic. Please do not take too much time to consider it.

Yes. This is a very difficult decision for me to make. I've had a lot of distractions recently (I am starting a PhD soon, and need to get a bunch of things in order beforehand) and put this decision off. I'll have an answer before this week is over, I promise!