Make the build's Ruby scripts run on Ruby 2.6 through 4.0

[dayglojesus]

Refs #83.

bin/gen_html spawned multimarkdown through Kernel#open with a leading |. Ruby 3.3 deprecated that (Feature #19630) and Ruby 4.0 removed it, so with a Homebrew Ruby 4 first on PATH every #!/usr/bin/env ruby build script that reaches gen_html fails, which is the About pages, the Help pages, the CHANGELOG page and the release notes. Nothing in the build hardcodes /usr/bin/ruby; the scripts just assumed an older Ruby.

Changes:

• gen_html uses IO.popen with an argv array.
• gen_html -o FILE writes next to the target and renames into place. The CMake rule uses it instead of a shell redirect. Before, a failed run left a truncated page that ninja then treated as up to date, so the broken page shipped in the app (seen while testing: a 295-byte About.html accepted by the next build).
• gen_credits.rb requires cgi/escape, the only part of cgi it uses and the only part Ruby 4.0 keeps. Ruby 4.0.7 still ships a stub cgi.rb that forwards to it, so this is spelling, not a breakage.
• gen_test initialises its buffer with String.new; Ruby 4 warns that the literal will be frozen.
• tests/gen_html_test.rb (minitest, run standalone like build_filetype_index_test.rb) covers the conversion, -o, and that a failed run leaves an existing output file untouched.

Verification on macOS 27.0 with the system Ruby 2.6.10 and Homebrew Ruby 4.0.7:

• All sixteen generated pages are byte-identical between the old script on 2.6 and the new script on 2.6 and on 4.0.7.
• ninja Applications/TextMate/md/About/About.html succeeds with Ruby 4.0.7 first on PATH; on main it fails at gen_html:97.
• tests/gen_html_test.rb: 3 runs, 13 assertions, 0 failures under each Ruby.
• gen_test output unchanged apart from the #line path; expand_variables and extract_changes were already fine on both.

[tbates]

Code review, all five files:

• textmate/bin/gen_html:97: open("|…", 'r+') → IO.popen([filter, "--nosmart"], 'r+') does exactly the right job here: Skips the shell and compatible with system 2.6 and future 4.0.

  • I ran the same IO.popen array + r+ + close_write pattern under system Ruby 2.6.10 with /bin/cat standing in for multimarkdown: round-trip passes.✓

• -o atomic write (tmp file + File.rename): good fix for the truncated-page problem, and stdout behavior is unchanged when -o is omitted.

  • One nit: a failed run orphans the .pid.tmp file; harmless, but a retry never cleans it?

• textmate/cmake/TextMateHelpers.cmake:206: switching the custom command to -o removes the shell-redirect dependency. Robust under the CMake generator.

• gen_credits.rb (cgi → cgi/escape): correct, CGI.escapeHTML lives in cgi/escape on both 2.6 and 4.0.

• gen_test ("" → String.new): harmless on 2.6, future-proofs frozen literals.

• tests/gen_html_test.rb: well designed, runs under whichever Ruby executes it via RbConfig.ruby, covers conversion, -o, and no-clobber-on-failure.

  • Note the file is standalone minitest, not wired into ctest (the t_*.cc glob will not pick up a .rb file), so CI must invoke ruby tests/gen_html_test.rb explicitly.

I scanned the other build Ruby scripts (expand_variables, extract_changes, show_log, update_changes, generate_available_bundles.rb) for pipe-open, exists?, Fixnum, URI.escape, untaint, bare cgi. All clean. The only other gen_html caller is release.yml:305, using stdout redirect, which keeps working.