c99 fixes

[flavorjones]

What problem is this PR intended to solve?

#2302 reports a compilation failure on RHEL6, RHEL7, and SLES12. These are older distros (RHEL7 is running gcc 4.8) which seem to compile in a C90-ish mode by default that gives errors on C99-isms.

This PR removed the C99isms (which are errors) as well as some C90isms (which generate warnings). Specifically:

• don't declare a variable in a for statement (C99-ism)
• don't mix declarations and code (C90-ism)

This PR also cleans up some C casting related to how we're handling const qualifiers. Specifically:

• gumbo.c is now using (const xmlChar *) in place of BAD_CAST (which is an alias for (xmlChar *))
• extracted the const-discard hack (xmlChar *)(uintptr_t)ptr into a macro DISCARD_CONST_QUAL
• introduce use of DISCARD_CONST_QUAL in one additional place in xml_node.c

As a result, even when using -Wcast-qual on older Rubies that support it, we now see no compiler warnings.

Have you included adequate test coverage?

I considered setting up CI using a comparable Centos docker image, but I didn't feel like there was much regression value to that test, given the nature of this failure. I could be argued into doing it, if anyone feels strongly.

Does this change affect the behavior of either the C or the Java implementations?

No functional changes.

[stevecheckoway]

FWIW, I think this is not a good change. C99 is now more than 20 years old. Unless we're really targeting a compiler that doesn't support it (and we can't be because gumbo itself makes heavy use of C99 features), we should figure out how to pass -std=c99 to the compiler.

(If there were features of C11 that I thought were useful, I'd say we should use -std=c11, but there really aren't for this.)

[stevecheckoway]

That should have referred only to 3e3485b and not the PR as a whole.

[stevecheckoway]

Does adding

append_cflags("-std=c99")

here not work?

[flavorjones]

Hi Steve, thanks for bringing this up, and I'm sorry I didn't wait for your input before whacking the "merge" button. I almost entirely agree with you despite having introduced the change.

But I will point out that @nobu has put effort into keeping the codebase of Ruby, Nokogiri, and other gems "clean" and removing C90-isms because (I think) Ruby 2.6 and earlier still use them, and I want to respect his wishes despite not fully understanding them. I've tagged Nobu here in case he has additional insight.

My only hesitation about passing -std=c99 or -std=c11 is that I have an incomplete understanding of compiler behavior, and I was worried that it might break compilation for some obscure platform. But your point about libgumbo using it is a good one and I'm thinking about whether to revert it. I'll prep that revert/c99 change in a separate PR and wait a bit to see if Nobu has an opinion.

[stevecheckoway]

Looks like -pedantic was removed from Ruby (by @nobu) in 2010 (although I don't claim to understand Ruby's build process).

It looks like -std=gnu99 was added some point before Ruby 2.6. But again, this area is confusing.

[flavorjones]

See #2304