[flang-dev] [RFC] Frontend driver: action vs non-action options
Andrzej Warzynski via flang-dev
flang-dev at lists.llvm.org
Thu Nov 4 01:55:21 PDT 2021
Let me just quickly summarise.
The behaviour of `flang-new -fc1` is consistent with `clang -cc1`, but
slightly different to Gfortran (in particular, with respect to the
`-fdump-parse-tree`). Also, the current implementation works as
expected, i.e. there are no known bugs that would require immediate
One could (and perhaps should?) refactor the dump options - currently
there's quite a few of them. There are a few options here -
`-fdump=unparse,symbols -o dump.txt` looks particularly promising.
I'm not sure whether any of Flang's dump options should be refactored as
regular flags, but it seems like the most natural way to replicate
GFortran's behaviour here. I won't block any efforts in this direction,
but please keep in mind that we will likely need Flang's parse tree for
future static analysis tools/frameworks (e.g. sanitizers?). This means
that the parse tree needs to be considered as regular compiler
input/output (as opposed to a plain debug dump).
Please let me know if you want to work on any of the above - I'll be
available for your questions and to review your patches. As for making
`flang-new -fc1` error out when multiple actions are specified, there's
https://reviews.llvm.org/D111781. I've reached out to cfe-dev to see
what they think , but there seemed to be little appetite for
discussion. Let me add some reviewers from LLVM instead. Pete has
already accepted it (thank you!), but since it touches both Flang *and*
LLVM, I want to make sure that developers from LLVM are also OK with this.
Thank you for all your feedback!
On 19/10/2021 19:31, Andrzej Warzynski via flang-dev wrote:
> We would like Flang's driver to be a drop-in replacement for e.g.
> `gfortran`. If we change this particular behaviour, we will be making
> the transition for our users harder.
More information about the flang-dev