`rb_str_resize(Qfalse, 0)` with NKF and without C extensions lock

[eregon]

Found by @mohamedhafez and reported on Slack:

class: TypeError
TruffleRuby doesn't have a case for the org.truffleruby.cext.CExtNodesFactory$RbStrResizeNodeFactory$RbStrResizeNodeGen node with values of type java.lang.Boolean=false java.lang.Integer=0
	from org.truffleruby.cext.CExtNodesFactory$RbStrResizeNodeFactory$RbStrResizeNodeGen.executeAndSpecialize(CExtNodesFactory.java:4234)
	from org.truffleruby.cext.CExtNodesFactory$RbStrResizeNodeFactory$RbStrResizeNodeGen.execute(CExtNodesFactory.java:4212)
	from org.truffleruby.language.RubyCoreMethodRootNode.execute(RubyCoreMethodRootNode.java:58)
string.c:197:in `rb_str_resize': TruffleRuby doesn't have a case for the org.truffleruby.cext.CExtNodesFactory$RbStrResizeNodeFactory$RbStrResizeNodeGen node with values of type java.lang.Boolean=false java.lang.Integer=0 (TypeError)
	from org.truffleruby.cext.CExtNodesFactory$RbStrResizeNodeFactory$RbStrResizeNodeGen.executeAndSpecialize(CExtNodesFactory.java:4234)
	from org.truffleruby.cext.CExtNodesFactory$RbStrResizeNodeFactory$RbStrResizeNodeGen.execute(CExtNodesFactory.java:4212)
	from org.truffleruby.language.RubyCoreMethodRootNode.execute(RubyCoreMethodRootNode.java:58)
	from string.c:197:in `rb_str_resize'
	from /Users/mohamed/.rbenv/versions/truffleruby+graalvm-24.0.1/lib/truffle/truffle/cext_ruby.rb:40:in `guess'
	from /Users/mohamed/rubyprojects/sa-mechanize/lib/mechanize/util.rb:58:in `detect_charset'
	from /Users/mohamed/rubyprojects/substitutealert/lib/ext/mechanize_additions.rb:22:in `initialize'

So this must be a call to rb_str_resize(str, 0) with str=0 but of course it shouldn't be 0.

Looking at NKF sources, there is a single call to rb_str_resize:

truffleruby/src/main/c/nkf/nkf.c

Line 47 in 938589f

rb_str_resize(result, o_len);

result is a global variable 😨

truffleruby/src/main/c/nkf/nkf.c

Line 40 in 938589f

static VALUE result;

truffleruby/src/main/c/nkf/nkf.c

Lines 164 to 168 in 938589f

	/* use _result_ begin*/
	result = tmp;
	kanji_convert(NULL);
	result = Qnil;
	/* use _result_ end */

So that is very likely the issue.
One solution is to use a native thread-local variable instead.
Or pass it as an extra parameter but that would require touching more of the upstream nkf code.

Or, always use the C extensions lock/a lock for NKF.

[andrykonchin]

It seems to me all the global variables could lead to such issues:

truffleruby/src/main/c/nkf/nkf.c

Lines 32 to 38 in 938589f

	static unsigned char *output;
	static unsigned char *input;
	static int input_ctr;
	static int i_len;
	static int output_ctr;
	static int o_len;
	static int incsize;

So all these variables probably should be passed everywhere (as a struct?). They are used only in this file so it doesn't seem a big effort to fix. Does it makes sense to do it?

[eregon]

The problem is e.g. rb_nkf_putchar() is a callback called as putchar() from src/main/c/nkf/nkf-utf8/nkf.c.
So I think it's too messy to pass a struct there.
Let's try using thread_local (https://en.cppreference.com/w/c/thread/thread_local) or __thread on these variables, that should require minimal changes.

[andrykonchin]

Created a PR ruby/nkf#19