perf(coverage): escape the HTML report once per file, not twice per line

[Chemaclass]

🤔 Background

Related #1096

An HTML report over this repo's src took 58.7 s for 128 files. Almost all
of it was one line: every source line was escaped through a command
substitution and a sed — roughly 22,000 processes for ~11,000 lines.

💡 Changes

• A whole file is escaped in one awk pass; 58.7 s → 9.5 s, output byte-identical over a fixed corpus (129 pages)
• The escaping deliberately stays out of Bash: 5.2 made a bare & in a substitution replacement mean the matched text, so ${line//</&lt;} yields <lt; there, while spelling it \& for 5.2 emits a literal backslash on 3.2 — verified on 3.2, 4.4 and 5.3, and both failures are silent
• Per-page wc -l | tr -d, three basename calls and a pwd become parameter expansions; the per-function percentage joins the class perf(coverage): the text report forks a subshell per file just to pick a colour #1092 converted, which had left one of the two forks behind
• A fork census pins the property that was fixed — sed count does not grow with source-line count — and reverting the emitter fails it

[Chemaclass]

🤔 Background

Related #1096

An HTML report over this repo's src took 58.7 s for 128 files. Almost all
of it was one line: every source line was escaped through a command
substitution and a sed — roughly 22,000 processes for ~11,000 lines.

💡 Changes

• A whole file is escaped in one awk pass; 58.7 s → 9.5 s, output byte-identical over a fixed corpus (129 pages)
• The escaping deliberately stays out of Bash: 5.2 made a bare & in a substitution replacement mean the matched text, so ${line//</&lt;} yields <lt; there, while spelling it \& for 5.2 emits a literal backslash on 3.2 — verified on 3.2, 4.4 and 5.3, and both failures are silent
• Per-page wc -l | tr -d, three basename calls and a pwd become parameter expansions; the per-function percentage joins the class perf(coverage): the text report forks a subshell per file just to pick a colour #1092 converted, which had left one of the two forks behind
• A fork census pins the property that was fixed — sed count does not grow with source-line count — and reverting the emitter fails it

[Chemaclass]

🤔 Background

Related #1096

An HTML report over this repo's src took 58.7 s for 128 files. Almost all
of it was one line: every source line was escaped through a command
substitution and a sed — roughly 22,000 processes for ~11,000 lines.

💡 Changes

• A whole file is escaped in one awk pass; 58.7 s → 9.5 s, output byte-identical over a fixed corpus (129 pages)
• The escaping deliberately stays out of Bash: 5.2 made a bare & in a substitution replacement mean the matched text, so ${line//</&lt;} yields <lt; there, while spelling it \& for 5.2 emits a literal backslash on 3.2 — verified on 3.2, 4.4 and 5.3, and both failures are silent
• Per-page wc -l | tr -d, three basename calls and a pwd become parameter expansions; the per-function percentage joins the class perf(coverage): the text report forks a subshell per file just to pick a colour #1092 converted, which had left one of the two forks behind
• A fork census pins the property that was fixed — sed count does not grow with source-line count — and reverting the emitter fails it

[Chemaclass]

🤔 Background

Related #1096

An HTML report over this repo's src took 58.7 s for 128 files. Almost all
of it was one line: every source line was escaped through a command
substitution and a sed — roughly 22,000 processes for ~11,000 lines.

💡 Changes

• A whole file is escaped in one awk pass; 58.7 s → 9.5 s, output byte-identical over a fixed corpus (129 pages)
• The escaping deliberately stays out of Bash: 5.2 made a bare & in a substitution replacement mean the matched text, so ${line//</&lt;} yields <lt; there, while spelling it \& for 5.2 emits a literal backslash on 3.2 — verified on 3.2, 4.4 and 5.3, and both failures are silent
• Per-page wc -l | tr -d, three basename calls and a pwd become parameter expansions; the per-function percentage joins the class perf(coverage): the text report forks a subshell per file just to pick a colour #1092 converted, which had left one of the two forks behind
• A fork census pins the property that was fixed — sed count does not grow with source-line count — and reverting the emitter fails it

[Chemaclass]

🤔 Background

Related #1096

An HTML report over this repo's src took 58.7 s for 128 files. Almost all
of it was one line: every source line was escaped through a command
substitution and a sed — roughly 22,000 processes for ~11,000 lines.

💡 Changes

• A whole file is escaped in one awk pass; 58.7 s → 9.5 s, output byte-identical over a fixed corpus (129 pages)
• The escaping deliberately stays out of Bash: 5.2 made a bare & in a substitution replacement mean the matched text, so ${line//</&lt;} yields <lt; there, while spelling it \& for 5.2 emits a literal backslash on 3.2 — verified on 3.2, 4.4 and 5.3, and both failures are silent
• Per-page wc -l | tr -d, three basename calls and a pwd become parameter expansions; the per-function percentage joins the class perf(coverage): the text report forks a subshell per file just to pick a colour #1092 converted, which had left one of the two forks behind
• A fork census pins the property that was fixed — sed count does not grow with source-line count — and reverting the emitter fails it

[Chemaclass]

🤔 Background

Related #1096

An HTML report over this repo's src took 58.7 s for 128 files. Almost all
of it was one line: every source line was escaped through a command
substitution and a sed — roughly 22,000 processes for ~11,000 lines.

💡 Changes

• A whole file is escaped in one awk pass; 58.7 s → 9.5 s, output byte-identical over a fixed corpus (129 pages)
• The escaping deliberately stays out of Bash: 5.2 made a bare & in a substitution replacement mean the matched text, so ${line//</&lt;} yields <lt; there, while spelling it \& for 5.2 emits a literal backslash on 3.2 — verified on 3.2, 4.4 and 5.3, and both failures are silent
• Per-page wc -l | tr -d, three basename calls and a pwd become parameter expansions; the per-function percentage joins the class perf(coverage): the text report forks a subshell per file just to pick a colour #1092 converted, which had left one of the two forks behind
• A fork census pins the property that was fixed — sed count does not grow with source-line count — and reverting the emitter fails it

[Chemaclass]

🤔 Background

Related #1096

An HTML report over this repo's src took 58.7 s for 128 files. Almost all
of it was one line: every source line was escaped through a command
substitution and a sed — roughly 22,000 processes for ~11,000 lines.

💡 Changes

• A whole file is escaped in one awk pass; 58.7 s → 9.5 s, output byte-identical over a fixed corpus (129 pages)
• The escaping deliberately stays out of Bash: 5.2 made a bare & in a substitution replacement mean the matched text, so ${line//</&lt;} yields <lt; there, while spelling it \& for 5.2 emits a literal backslash on 3.2 — verified on 3.2, 4.4 and 5.3, and both failures are silent
• Per-page wc -l | tr -d, three basename calls and a pwd become parameter expansions; the per-function percentage joins the class perf(coverage): the text report forks a subshell per file just to pick a colour #1092 converted, which had left one of the two forks behind
• A fork census pins the property that was fixed — sed count does not grow with source-line count — and reverting the emitter fails it

[Chemaclass]

🤔 Background

Related #1096

An HTML report over this repo's src took 58.7 s for 128 files. Almost all
of it was one line: every source line was escaped through a command
substitution and a sed — roughly 22,000 processes for ~11,000 lines.

💡 Changes

• A whole file is escaped in one awk pass; 58.7 s → 9.5 s, output byte-identical over a fixed corpus (129 pages)
• The escaping deliberately stays out of Bash: 5.2 made a bare & in a substitution replacement mean the matched text, so ${line//</&lt;} yields <lt; there, while spelling it \& for 5.2 emits a literal backslash on 3.2 — verified on 3.2, 4.4 and 5.3, and both failures are silent
• Per-page wc -l | tr -d, three basename calls and a pwd become parameter expansions; the per-function percentage joins the class perf(coverage): the text report forks a subshell per file just to pick a colour #1092 converted, which had left one of the two forks behind
• A fork census pins the property that was fixed — sed count does not grow with source-line count — and reverting the emitter fails it