readability-else-after-return only warns about return calls, but Coding Stadnards mention throw, continue, goto, etc.
Details
Diff Detail
- Repository
- rL LLVM
Event Timeline
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
38–40 ↗ | (On Diff #67160) | I think it might be better to specifically identify the flow control construct used instead of a laundry list of possibilities. Then it can be "do not use '%select{else|else if}0' after '%1'". |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
38–40 ↗ | (On Diff #67160) | The wording is actually good. For documentation (currently, http://clang.llvm.org/extra/clang-tidy/checks/readability-else-after-return.html is not extremely verbose ;)). Otherwise, I agree with Aaron: the warning should be precise to avoid confusion. Also, no need for a separate variable to keep the message, IMO. |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
25 ↗ | (On Diff #67347) | Let's structure this differently to reduce code duplication: auto returnLikeStmt = stmt(anyOf(returnStmt().bind("return"), continueStmt().bind("continue"), ...)); Finder->addMatcher(stmt(forEach(ifStmt(hasThen(anyOf(returnLikeStmt, compoundStmt(has(returnLikeStmt)))), hasElse(... |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
45 ↗ | (On Diff #67347) | This won't work in MSVC2013, I think. Just add a const char *Labels[] = {"return", ... (add more consts, constexprs or statics, if you like ;) |
51 ↗ | (On Diff #67347) | Please use diagnostic formatting facilities instead of string concatenation: diag(L, "xxx '%0' yyy") << ControlFlowInterruptor; |
docs/clang-tidy/checks/readability-else-after-return.rst | ||
10 ↗ | (On Diff #67347) | I would omit goto, since it's unclear, where the target label is (and it's not a common construct anyway). |
test/clang-tidy/readability-else-after-return.cpp | ||
33 ↗ | (On Diff #67347) | CHECK-FIXES pattern should be sticter. Current pattern will match for any // comment occurrence anywhere in the test file, even if it's on a different line and still prefixed with else. Two action items here: 1. make comments and patterns unique, 2. anchor patterns to the start of line ({{ˆ}}). |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
45 ↗ | (On Diff #67347) | Hm, I'm confused. What exactly wouldn't work in MSVC? |
test/clang-tidy/readability-else-after-return.cpp | ||
33 ↗ | (On Diff #67347) | Why make comments and patterns unique? If I anchor patterns to the start of line, it will only look at the line above, why does it need to be unique then? |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
45 ↗ | (On Diff #67347) | This one *will* work in MSVC 2013 (it was a different initializer list that @alexfh was thinking of, perhaps, where the deduced type of the list was a bit more complex). I just tried it out and the above construct compiles fine with MSVC. That being said, const auto & is the wrong type here, it should be const auto *. |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
45 ↗ | (On Diff #67347) | Hm, I'm confused. Why should it be a pointer? So, when I have something like: std::vector<unsigned> vec = {1, 2, 3}; for (const auto &x : vec) ... This is right, I guess. Why is that different with init list? |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
45 ↗ | (On Diff #67347) | The deduced type for the code as-written is const char * const &, which isn't wrong, it's just not particularly useful. If you switch to const auto *, the deduced type is the much more reasonable const char *. |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
45 ↗ | (On Diff #67347) | Ah... Thank you very much! Now I get it. I thought for whatever reason that there is implicit conversion to std::string. |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
45 ↗ | (On Diff #67347) |
With MSVC you never know what will work, and what won't... Thanks for verifying! |
test/clang-tidy/readability-else-after-return.cpp | ||
---|---|---|
33 ↗ | (On Diff #67347) |
Because FileCheck is way simpler than one usually thinks of it. It doesn't maintain any relation between the location of the check line and the input line. So your assumption that
is wrong. The only thing FileCheck cares about is that the input contains a subsequence of lines that match all the CHECK patterns (a bit more complicated, if CHECK-NOT, CHECK-NEXT, CHECK-SAME, CHECK-DAG, or CHECK-LABEL are in use, but the basic idea is roughly the same). So, if all CHECK-FIXES patterns in this test are // comment, then any input that contains at least the same number of // comment substrings on different lines will pass the test. Please read http://llvm.org/docs/CommandGuide/FileCheck.html as well. |
docs/clang-tidy/checks/readability-else-after-return.rst | ||
---|---|---|
8 ↗ | (On Diff #67476) | enforcements (plural) |
10 ↗ | (On Diff #67476) | I would remove the "etc", since this check does not handle such cases. |
12 ↗ | (On Diff #67476) | Following -> The following |
14 ↗ | (On Diff #67476) | Since we're talking about LLVM coding conventions, can you format this like LLVM code (elide braces, etc)? |
14 ↗ | (On Diff #67476) | This is not the proper way to introduce a code block. Should use .. code-block:: c++ and then indent the code blocks. |
clang-tidy/readability/ElseAfterReturnCheck.cpp | ||
---|---|---|
40 ↗ | (On Diff #67476) | No braces around single-line if bodies, please. Clang-tidy code is mostly consistent wrt this. |
docs/clang-tidy/checks/readability-else-after-return.rst | ||
14 ↗ | (On Diff #67476) | I think, the braces are fine, since the example needs to demonstrate that the check is able to remove them. |
docs/clang-tidy/checks/readability-else-after-return.rst | ||
---|---|---|
14 ↗ | (On Diff #67476) | Okay, that's a fair, but subtle, reason to leave them in. |