[clang] [Clang][Driver][Test] Created test for unsupported driver options (PR #120900)

[via cfe-commits]

GeorgeKA wrote:

@Maetveis 
Do you have any thoughts on the following two groups of options that are unexpectedly succeeding?
I'll note that I did come across some options for which the driver appears to correct user error by adding the appropriate driver option, like testing "`-fexperimental-sanitize-metadata=atomics`" with the default driver and the driver adding `-cc1`, but I'm not aware of a supported `-cc1` option that outputs the final list of options used like `-###` does for the default driver.

1) "`fheinous-gnu-extensions`" & "`fcuda-approx-transcendentals`" with `-cc1`
Both of these unexpectedly succeed with `clang -cc1`. Notably, they both have aliases that are supported for `-cc1`, ([ex: fheinous-gnu-extensions](https://github.com/llvm/llvm-project/blob/75781fa3c77f14640459a97d4265dda76d80e453/clang/include/clang/Driver/Options.td#L2823)), but those aliases expectedly fail when used with `-cc1`, which suggests aliases aren't the issue.

Ex: `clang -cc1 -fheinous-gnu-extensions -x c++ - < /dev/null 2>&1`

2) "`mno-strict-align`" & "`mstrict-align`" with `-cc1`
Both of these unexpectedly succeed with `clang -cc1`. Both of these have no alias. They're part of the m_Group and the CompileOnly_Group, which don't include `-cc1` for visibility.

Ex: `clang -cc1 -mno-strict-align -x c++ - < /dev/null 2>&1`

https://github.com/llvm/llvm-project/pull/120900