The list of boolean flags and others is becoming unresonnably long.
Details
Diff Detail
- Repository
- rG LLVM Github Monorepo
Event Timeline
mlir/include/mlir/Tools/mlir-opt/MlirOptMain.h | ||
---|---|---|
38 | Should these have default true ? E..g., config.setSplitInputFile(); Reads naturally as enabling to me (and I believe we follow that style with printer). | |
118 | Should we add the [[deprecated]] marker so that we get warnings? | |
146 | Not quite sure i follow the ordering here (at least among the bools) |
mlir/include/mlir/Tools/mlir-opt/MlirOptMain.h | ||
---|---|---|
38 | Ah right, I set I shouldn't use set for the fluent style API but use enable or use depending on the flag: config.enableSplitInputFile() | |
118 | We should, but can't hear, we have in-tree users. I actually plan to kill this soon (I'll put it on the new deprecations page on the website :)). | |
146 | I think I took the order the cl::opt are defined in the source (it's a bad reason, it's just that I had split screen editor and went one by one!). Should I put them in alphabetical order? |
mlir/include/mlir/Tools/mlir-opt/MlirOptMain.h | ||
---|---|---|
59–60 | the docstrings on a few of these are wrong, seems the setVerifyPasses comment got copied |
C:\ws\w4\llvm-project\premerge-checks\mlir\include\mlir/Tools/mlir-opt/MlirOptMain.h(65): error C2039: 'function': is not a member of 'std'
(Will fix windows, it is still on my radar)
Should these have default true ? E..g.,
config.setSplitInputFile();
Reads naturally as enabling to me (and I believe we follow that style with printer).