extconf not adding an appropriate -I for gumbo-parser on macOS

[stevecheckoway]

Please describe the bug

When compiling the C extension on macOS, compiling gumbo.c fails because it cannot find nokogiri_gumbo.h. The real problem is that the -Ipath/to/gumbo-parser/include is not getting passed to the compiler.

Here's the relevant line in the mkmf.log file

block in append_cppflags: checking for whether -I/Users/steve/programming/nokogiri/tmp/x86_64-darwin21/nokogiri/3.1.2/ports/x86_64-darwin21/libgumbo/1.0.0-nokogiri/include is accepted as CPPFLAGS... -------------------- no

The underlying problem seems to be that the configuration test is using -Werror along with the unsupported warning -Wmisleading-indentation.

Here's the relevant portion of the log. Note that checking for the flag fails but that the last line shows checking for gumbo_parse_with_options() succeeds.

block in append_cppflags: checking for whether -I/Users/steve/programming/nokogiri/tmp/x86_64-darwin21/nokogiri/3.1.2/ports/x86_64-darwin21/libgumbo/1.0.0-nokogiri/include is accepted as CPPFLAGS... -------------------- no

DYLD_FALLBACK_LIBRARY_PATH=.:/nix/store/dmg6nslgajiyw8gqm59yc545yvg8axqx-ruby-3.1.2/lib "clang -I/nix/store/dmg6nslgajiyw8gqm59yc545yvg8axqx-ruby-3.1.2/include/ruby-3.1.0/x86_64-darwin21 -I/nix/store/dmg6nslgajiyw8gqm59yc545yvg8axqx-ruby-3.1.2/include/ruby-3.1.0/ruby/backward -I/nix/store/dmg6nslgajiyw8gqm59yc545yvg8axqx-ruby-3.1.2/include/ruby-3.1.0 -I../../../../ext/nokogiri -I/Users/steve/programming/nokogiri/ports/x86_64-darwin21/libxml2/2.10.3/include/libxml2 -I/Users/steve/programming/nokogiri/ports/x86_64-darwin21/libxslt/1.1.37/include -I/Users/steve/programming/nokogiri/ports/x86_64-darwin21/libxml2/2.10.3/include/libxml2 -D_XOPEN_SOURCE -D_DARWIN_C_SOURCE -D_DARWIN_UNLIMITED_SELECT -D_REENTRANT   -fdeclspec -O3 -fno-fast-math -ggdb3 -Wall -Wextra -Wdeprecated-declarations -Wdivision-by-zero -Wimplicit-function-declaration -Wimplicit-int -Wmisleading-indentation -Wpointer-arith -Wshorten-64-to-32 -Wwrite-strings -Wold-style-definition -Wmissing-noreturn -Wno-constant-logical-operand -Wno-long-long -Wno-missing-field-initializers -Wno-overlength-strings -Wno-parentheses-equality -Wno-self-assign -Wno-tautological-compare -Wno-unused-parameter -Wno-unused-value -Wunused-variable -Wextra-tokens -Wundef  -fno-common -pipe  -I/Users/steve/programming/nokogiri/tmp/x86_64-darwin21/nokogiri/3.1.2/ports/x86_64-darwin21/libgumbo/1.0.0-nokogiri/include -Werror -c conftest.c"
error: unknown warning option '-Wmisleading-indentation'; did you mean '-Wbinding-in-condition'? [-Werror,-Wunknown-warning-option]
checked program was:
/* begin */
1: #include "ruby.h"
2: 
3: int main(int argc, char **argv)
4: {
5:   return !!argv[argc];
6: }
/* end */

--------------------

have_func: checking for gumbo_parse_with_options() in nokogiri_gumbo.h... -------------------- yes

One work-around is to add

append_cflags("-Wno-unknown-warning-option") if darwin?

to extconf.rb. But maybe a better option would be to figure out where -Wmisleading-indentation (or-Werror) is coming from and fix that.

[flavorjones]

@stevecheckoway Likely that compiler flag is coming from Ruby itself. Can you take a look at the value of RbConfig::CONFIG["CFLAGS"] and see if it's there?

If so, one possible cause is that you've upgraded clang or xcode or whatever since that Ruby executable was built, and so your current compiler doesn't understand flags that were used to compile Ruby with your previous compiler.

[stevecheckoway]

Good thinking! I suspect that what happened is my Ruby was compiled by a newer compiler (via Nix) than what I have installed on that machine. (That particular machine is stuck on the last version of macOS that supports 32-bit binaries.) I'll head into the office (where that machine is located currently) and check.

[flavorjones]

Regardless, it would be good to add that clang flag on darwin. Can you kick the tires on #2690?

[stevecheckoway]

#2690 fixes the problem. You were correct about the cause:

irb(main):002:0> RbConfig::CONFIG["CFLAGS"].split
=>
["-fdeclspec",
 "-O3",
 "-fno-fast-math",
 "-ggdb3",
 "-Wall",
 "-Wextra",
 "-Wdeprecated-declarations",
 "-Wdivision-by-zero",
 "-Wimplicit-function-declaration",
 "-Wimplicit-int",
 "-Wmisleading-indentation",        # <---
 "-Wpointer-arith",
 "-Wshorten-64-to-32",
 "-Wwrite-strings",
 "-Wold-style-definition",
 "-Wmissing-noreturn",
 "-Wno-constant-logical-operand",
 "-Wno-long-long",
 "-Wno-missing-field-initializers",
 "-Wno-overlength-strings",
 "-Wno-parentheses-equality",
 "-Wno-self-assign",
 "-Wno-tautological-compare",
 "-Wno-unused-parameter",
 "-Wno-unused-value",
 "-Wunused-variable",
 "-Wextra-tokens",
 "-Wundef",
 "-fno-common",
 "-pipe"]

[flavorjones]

👍 when that PR goes green I'll merge it. In the meantime you should be able to work around this by setting your environment variable CFLAGS because we do this in the extconf:

append_cflags(ENV["CFLAGS"].split) unless ENV["CFLAGS"].nil?