Disallow RETURNDATASIZE and RETURNDATACOPY in inline assembly blocks in pure functions

[matheusaaguiar]

(cherry picked from commit f567eb1)

This PR recreates #12861 on breaking, after it was reverted from develop in #13013.

[cameel]

Two of the CI failures seem to be coming from other PRs. For example negation.sol:

Expected result:
  TypeError 4907: (97-99): Unary operator - cannot be applied to type uint256: Unary negation is only allowed for signed integers.
Obtained result:
  TypeError 4907: (97-99): Unary operator - cannot be applied to type uint256. Unary negation is only allowed for signed integers.

This really weird. The test was added in #12828 and I remember that that PR had a last-minute change to replace the : with .. But the version with : was just changed in place. It was never merged into develop or breaking. Now this PR behaves as if it had :.

What'e ever weirder, I just checked state of TypeChecker.cpp as of this PR and there's clearly . there:

solidity/libsolidity/analysis/TypeChecker.cpp

Line 1665 in 932fb3e

description += ". " + result.message();

So it's as if the binary used to run tests here was not built from the code at the right commit.

[cameel]

@matheusaaguiar I updated the description to link to the old PR. If your PR is directly related to some other PRs please try to always link to them. It helps a lot if, for whatever reason, you later need to go back and figure out what actually happened.

Also when github folds part of the title into the description, just put it back in the title. Or shorten it - whichever you prefer.

[cameel]

I took a closer look at history and I think that the extra failures are actual errors on breaking that probably have not been fixed after the last merge from develop.

The negation.sol change for example:

• That change in TypeChecker.cpp was originally done on breaking in Disallow delete on types containing nested mappings. #11843. Back then it had : in the message.
• Recently the same change was introduced on develop in Use error message for unary operators. #12828. I thought it was something completely new but now I realized that it must have just been cherry-picked from breaking.
  • I suggested a small tweak in that PR: we changed : to . as a fixup. That's why I was confused as to why : is still there. I thought that PR was the only thing changing that code and that we had : only in its initial version.
• Later develop was merged into breaking and there must have been a conflict due to that tweak. I think it must have been resolved by taking TypeChecker.cpp from one PR but negation.sol from the other. Since such a merge usually triggers a wave of fail notifications, the fact that breaking was broken because of that was probably not noticed.

So, in short, these seem to be actual problems on breaking and should be fixed in a separate PR to breaking.

[ekpyron]

Since we said we want to have @wechman fix the test failures due to merging develop into breaking separately, I now merged #13041 - but this PR should now be fine again by rebasing and just dropping the test fixes that are unrelated to this PR from here, since they're now already in breaking.

[ekpyron]

Looks good!

[cameel]

Looks good.

I'm going to merge this despite the failing ext test since it's unrelated and it might take some time before the fix gets merged into breaking.