First, I updated the clang rewriter to do the conversion.
Then I lightly hand-tweaked the output for the few places where
the rewriter messed up, mostly when dealing with calls on "this".
Change-Id: I40a6a977959cd97415c678eafc8507de8aa3b1a9
Reviewed-on: https://gerrit.libreoffice.org/7879
Reviewed-by: Michael Stahl <mstahl@redhat.com>
Tested-by: Michael Stahl <mstahl@redhat.com>
...which warns about apparent errors like
n == KIND_FOO || KIND_BAR
that should have read
n == KIND_FOO || n == KIND_BAR
It also warns about trivial (mis-)uses of 0/1 and sal_False/True like
while (1) { ... }
instead of
while (true) { ... }
which it can automatically rewrite.
It does not warn if the relevant code is inside a macro in a .h (i.e., C)
include file, nor when a string literal is used in the assert macro like
assert(!"this cannot happen");
Change-Id: If8c305a25e5da15e78f6139b5b0c0e868f06d4f3
...which can act as either a rewriter or a non-rewriter that emits warnings.
Also added COMPILER_PLUGIN_WARNINGS_ONLY=X to demote warnings from plugin X from
errors to warnings, even under --enable-werror.
Change-Id: I05361936240a890515c6bba2459565417c1746b7
...as MSVC would warn about those anyway (at least about cases that do not
compare against literal sal_True/sal_False, but warning even about those helped
clean up lots of redundant
if (foo == true)
instead of just
if (foo)
etc. across the code base).
Change-Id: Iad4b335c35c5411070f04f87f974db7942c288d4
...as they are often enough errors, like a typo in brace placement in
if (foo == (FOO || bar == BAR) && baz)
or a literal true passed as an argument to a function that rather expects an
integer bit mask, etc. The plugin is smart enough to detect interaction with
logically boolean return/parameter types of C functions that use [unsigned] int
instead, and knows the relevant boolean typedefs (sal_Bool, gboolean, etc.).
Change-Id: I5f0e4344fe86589bec35a71018c7effdedf85e3e
Use a clang rewriter to rewrite SvStream::operator<< to methods
like WriteUInt32.
Note that the rewriter is not perfect, and hand-tweaking the output
is necessary.
Change-Id: I0291c8192ca74d6334ed3cf8cb713212b2f0c67d
Reviewed-on: https://gerrit.libreoffice.org/7307
Reviewed-by: Michael Stahl <mstahl@redhat.com>
Tested-by: Michael Stahl <mstahl@redhat.com>
... it is an abbreviation of "Solar Version".
Since nobody can remember that:
remove OUTDIR OUTDIR_FOR_BUILD SOLARVER SOLARVERSION solarpath
and any mention thereof.
Change-Id: Idb3031c4f25a76ac05b22ec67e3ca3e1e8e512ad
Reviewed-on: https://gerrit.libreoffice.org/6515
Tested-by: LibreOffice gerrit bot <gerrit@libreoffice.org>
Reviewed-by: Michael Stahl <mstahl@redhat.com>
Compiler plugin to replace with matching number(), boolean() or OUString ctor,
ran it, few manual tweaks, mark as really deprecated.
Change-Id: I4a79bdbcf4c460d21e73b635d2bd3725c22876b2
Apparently Clang rewriter API doesn't properly bail out if the range
given is not inside one file, so check explicitly.
Change-Id: I27be6d396a131d385231e9c0dfa8c84d9fa15ccc
Given that locations often point to a (start of) token, even if it's
e.g. getLocEnd(), this should be very useful.
Change-Id: I266e4c0a234262e99158c8f495b631f54f8a5608
Some improvements, like making it simple to actually remove a statement
or a token including its associated whitespace.
Change-Id: I02a5bd919f1fadae1dcd45a76f9d25df353ac518
The warn_unused attribute has been upstream to GCC and Clang, so use it if present.
Still warn about STL types if those do not use it yet (which is the status as of now).
Change-Id: I3c003e44c08d1d141e23bba38cf92e663a5ac353
C/C++ silently converts pointers to booleans, which is presumably seen as
a good idea by lazy people, but it can occassionally silently break code
in strange ways, most notably by selecting a bool overload when no matching
overload for the pointer exists (OUStringBuffer::append() can break like that,
e.g. in ba37e4062f538db7e51d6a64ba544eeddbc567cf, other cases are
8e3bf1598fa95ac8d099e45ae4252e7654a6f590 or 28e4c0250e67a344b4d6088bdca2e680a4bffad0).
So far the plugin checks only conversions in function call arguments, there's
disabled code that could check more places, but I'm not aware so far of any place
where anything actually broke because of that, and it'd require fixups to be
explicit in some places (e.g. 'bool b = returns_pointer();' would require '!= NULL'
added), so for it'll be only the simple various and it can be made more strict if wanted.
Change-Id: I6a5d207daf925e6c2d1bf684060536795ecfcc35
Ctor bodies can also have code inside of member variables initialization,
which is not considered to be inside function body.
Change-Id: Id68960093a51396b9486f1364b1a361526c3431d
Clang API doesn't provide this, but it's occasionally needed, and so far
the way has been inspecting the highest possible node in AST and walking down
and remembering, which is complicated, error-prone and annoying.
Change-Id: Id5b72cb5ebfc069e90efe6d673c0ef18ebcdab61
A different way to do 1c0669af2f1f58e6431b5e489ac48a883e242ba7.
Sometimes one piece of code can be represented several times in the AST,
e.g. with default function arguments.
Change-Id: Ic7799fa0bd918a638bdc8ebef69e6aa91d355bdc
This does not always work well, e.g. when building a return value
in a return statement from a temporary, there is CXXConstructExpr
containing CXXTemporaryObjectExpr, which both share the same location.
This reverts commit 1c0669af2f1f58e6431b5e489ac48a883e242ba7.
