clang-format: Disable across-block include reordering.

Differential Revision: https://reviews.llvm.org/D66650

llvm-svn: 369759

1 files changed, 21 insertions, 0 deletions

diff --git a/clang/lib/Format/Format.cpp b/clang/lib/Format/Format.cpp
index 175807edfb4..90745b6e0a6 100644
--- a/clang/lib/Format/Format.cpp
+++ b/clang/lib/Format/Format.cpp
@@ -910,6 +910,27 @@ FormatStyle getGoogleStyle(FormatStyle::LanguageKind Language) {
 
 FormatStyle getChromiumStyle(FormatStyle::LanguageKind Language) {
   FormatStyle ChromiumStyle = getGoogleStyle(Language);
+
+  // Disable include reordering across blocks in Chromium code.
+  // - clang-format tries to detect that foo.h is the "main" header for
+  //   foo.cc and foo_unittest.cc via IncludeIsMainRegex. However, Chromium
+  //   uses many other suffices (_win.cc, _mac.mm, _posix.cc, _browsertest.cc,
+  //   _private.cc, _impl.cc etc) in different permutations
+  //   (_win_browsertest.cc) so disable this until IncludeIsMainRegex has a
+  //   better default for Chromium code.
+  // - The default for .cc and .mm files is different (r357695) for Google style
+  //   for the same reason. The plan is to unify this again once the main
+  //   header detection works for Google's ObjC code, but this hasn't happened
+  //   yet. Since Chromium has some ObjC code, switching Chromium is blocked
+  //   on that.
+  // - Finally, "If include reordering is harmful, put things in different
+  //   blocks to prevent it" has been a recommendation for a long time that
+  //   people are used to. We'll need a dev education push to change this to
+  //   "If include reordering is harmful, put things in a different block and
+  //   _prepend that with a comment_ to prevent it" before changing behavior.
+  ChromiumStyle.IncludeStyle.IncludeBlocks =
+      tooling::IncludeStyle::IBS_Preserve;
+
   if (Language == FormatStyle::LK_Java) {
     ChromiumStyle.AllowShortIfStatementsOnASingleLine =
         FormatStyle::SIS_WithoutElse;