ALE appends flags from {c,cpp}_{clang,gcc}_options after those found by
parsing compile_commands.json or Makefile output. If -std=* flags are
present in both the ALE flags and parsed flags, the last one present
(i.e., ALE's -std=* flag) will determine the mode the compiler works in.
This can result in errors showing up in vim but not in the actual build
or vice-versa.
For example, say you have foo.cpp:
#include <type_traits>
int main() {
return std::is_same_v<float, int>;
}
If cpp_clang_options contains -std=c++17 and -std=c++14 is parsed from
compile_commands.json, then ALE would end up running something like:
clang++ -S -x c++ -fsyntax-only -std=c++14 -std=c++17 - < foo.cpp
This would result in no errors showing up in Vim, but the actual build
would fail with:
<stdin>:3:14: error: no template named 'is_same_v' in namespace 'std'; did you mean 'is_same'?
return std::is_same_v<float, int>;
~~~~~^~~~~~~~~
is_same
/Library/Developer/CommandLineTools/usr/bin/../include/c++/v1/type_traits:872:61: note: 'is_same' declared here
template <class _Tp, class _Up> struct _LIBCPP_TEMPLATE_VIS is_same : public false_type {};
^
<stdin>:3:35: error: expected '(' for function-style cast or type construction
return std::is_same_v<float, int>;
~~~~~~~~~~~~~~~~~~~~~~~~~~^
2 errors generated.
as the actual build would not have the -std=c++17 flag added by ALE.
If cpp_clang_options contains -std=c++14 and -std=c++17 is parsed from
compile_commands.json, then the opposite problem would occur. ALE would
end up running something like:
clang++ -S -x c++ -fsyntax-only -std=c++17 -std=c++14 - < foo.cpp
and would show an error on line 3 of foo.cpp:
[clang] No template named 'is_same_v' in namespace 'std'; did you mean 'is_same'? (fix available)
The actual build, on the other hand, would succeed without any
complaints.
Removing -std=* from ALE's flags if it is already present in the parsed
flags ensures that the wrong -std=* flag is not used.
An alternative would have been to switch the order in which parsed flags
and ALE flags were concatenated when producing the command to execute,
but that could prevent a user from intentionally using ALE's flags to
override some other flags, e.g. -W* flags to enable/disable warnings in
a project whose flags are not under the developer's control.
-std=* flags are also present in cuda/nvcc.vim, objc/clang.vim,
objcpp/clang.vim, and vhdl/ghdl.vim, but none of those linters appear to
parse compile_commands.json or `make` output.
This PR first and formost implements support for dot-seperate filetypes,
a very trivial change.
This closes#132
But more importantly, this PR vastly improves the test quality for
`ale#linter#Get`. It enables us to reset the state of ale's internal
linter cache, to facilitate better testing, as well as making use of
mocked linters instead of depending on linters on disk (which may
change). In addition, a dummy linter is defined to test the autoloading
behavior.
Header guards were removed from all linters as:
* A: ale won't try and load linters if they already exist in memory
* B: we can't reset state for testing if they can't be loaded again
* First pass at optimizing ale to autoload
First off, the structure/function names should be revised a bit,
but I will wait for @w0rp's input before unifying the naming style.
Second off, the docs probably need some more work, I just did some
simple find-and-replace work.
With that said, this pull brings major performance gains for ale. On my
slowest system, fully loading ale and all its code takes around 150ms.
I have moved all of ale's autoload-able code to autoload/, and in
addition, implemented lazy-loading of linters. This brings load time on
that same system down to 5ms.
The only downside of lazy loading is that `g:ale_linters` cannot be
changed at runtime; however, it also speeds up performance at runtime by
simplfying the logic greatly.
Please let me know what you think!
Closes#59
* Address Travis/Vint errors
For some reason, ale isn't running vint for me...
* Incorporate feedback, make fixes
Lazy-loading logic is much improved.
* Add header comments; remove incorrect workaround
* Remove unneeded plugin guards
* Fix lazy-loading linter logic
Set the wrong variable....
* Fix capitialization
