perf(coverage): emit each HTML page's code table in one awk pass

[Chemaclass]

🤔 Background

Related #1098

After the escaping fix, generate_file_html was 8.1 s of a 9.5 s report: a
Bash loop classifying, looking up and echoing markup for each of 22,405 source
lines, plus 15 cat <<EOF blocks per page — 1935 forks for a 129-page report.

💡 Changes

• The code table is one awk pass, reading the hits block, the per-line test list and the source together
• Markup blocks go through a builtin read loop instead of cat
• 9.5 s → 4.5 s, or 58.7 s → 4.5 s counting perf(coverage): escape the HTML report once per file, not twice per line #1097
• Byte-identical on a fixed corpus, verified both with no hits and with real hits and tooltips — the no-hits corpus never renders a badge, so it alone would not catch a mistake there
• The badge's × is passed as an awk variable rather than \xc3\x97, since \x is not POSIX awk

[Chemaclass]

🤔 Background

Related #1098

After the escaping fix, generate_file_html was 8.1 s of a 9.5 s report: a
Bash loop classifying, looking up and echoing markup for each of 22,405 source
lines, plus 15 cat <<EOF blocks per page — 1935 forks for a 129-page report.

💡 Changes

• The code table is one awk pass, reading the hits block, the per-line test list and the source together
• Markup blocks go through a builtin read loop instead of cat
• 9.5 s → 4.5 s, or 58.7 s → 4.5 s counting perf(coverage): escape the HTML report once per file, not twice per line #1097
• Byte-identical on a fixed corpus, verified both with no hits and with real hits and tooltips — the no-hits corpus never renders a badge, so it alone would not catch a mistake there
• The badge's × is passed as an awk variable rather than \xc3\x97, since \x is not POSIX awk

[Chemaclass]

🤔 Background

Related #1098

After the escaping fix, generate_file_html was 8.1 s of a 9.5 s report: a
Bash loop classifying, looking up and echoing markup for each of 22,405 source
lines, plus 15 cat <<EOF blocks per page — 1935 forks for a 129-page report.

💡 Changes

• The code table is one awk pass, reading the hits block, the per-line test list and the source together
• Markup blocks go through a builtin read loop instead of cat
• 9.5 s → 4.5 s, or 58.7 s → 4.5 s counting perf(coverage): escape the HTML report once per file, not twice per line #1097
• Byte-identical on a fixed corpus, verified both with no hits and with real hits and tooltips — the no-hits corpus never renders a badge, so it alone would not catch a mistake there
• The badge's × is passed as an awk variable rather than \xc3\x97, since \x is not POSIX awk

[Chemaclass]

🤔 Background

Related #1098

After the escaping fix, generate_file_html was 8.1 s of a 9.5 s report: a
Bash loop classifying, looking up and echoing markup for each of 22,405 source
lines, plus 15 cat <<EOF blocks per page — 1935 forks for a 129-page report.

💡 Changes

• The code table is one awk pass, reading the hits block, the per-line test list and the source together
• Markup blocks go through a builtin read loop instead of cat
• 9.5 s → 4.5 s, or 58.7 s → 4.5 s counting perf(coverage): escape the HTML report once per file, not twice per line #1097
• Byte-identical on a fixed corpus, verified both with no hits and with real hits and tooltips — the no-hits corpus never renders a badge, so it alone would not catch a mistake there
• The badge's × is passed as an awk variable rather than \xc3\x97, since \x is not POSIX awk

[Chemaclass]

🤔 Background

Related #1098

After the escaping fix, generate_file_html was 8.1 s of a 9.5 s report: a
Bash loop classifying, looking up and echoing markup for each of 22,405 source
lines, plus 15 cat <<EOF blocks per page — 1935 forks for a 129-page report.

💡 Changes

• The code table is one awk pass, reading the hits block, the per-line test list and the source together
• Markup blocks go through a builtin read loop instead of cat
• 9.5 s → 4.5 s, or 58.7 s → 4.5 s counting perf(coverage): escape the HTML report once per file, not twice per line #1097
• Byte-identical on a fixed corpus, verified both with no hits and with real hits and tooltips — the no-hits corpus never renders a badge, so it alone would not catch a mistake there
• The badge's × is passed as an awk variable rather than \xc3\x97, since \x is not POSIX awk

[Chemaclass]

🤔 Background

Related #1098

After the escaping fix, generate_file_html was 8.1 s of a 9.5 s report: a
Bash loop classifying, looking up and echoing markup for each of 22,405 source
lines, plus 15 cat <<EOF blocks per page — 1935 forks for a 129-page report.

💡 Changes

• The code table is one awk pass, reading the hits block, the per-line test list and the source together
• Markup blocks go through a builtin read loop instead of cat
• 9.5 s → 4.5 s, or 58.7 s → 4.5 s counting perf(coverage): escape the HTML report once per file, not twice per line #1097
• Byte-identical on a fixed corpus, verified both with no hits and with real hits and tooltips — the no-hits corpus never renders a badge, so it alone would not catch a mistake there
• The badge's × is passed as an awk variable rather than \xc3\x97, since \x is not POSIX awk

[Chemaclass]

🤔 Background

Related #1098

After the escaping fix, generate_file_html was 8.1 s of a 9.5 s report: a
Bash loop classifying, looking up and echoing markup for each of 22,405 source
lines, plus 15 cat <<EOF blocks per page — 1935 forks for a 129-page report.

💡 Changes

• The code table is one awk pass, reading the hits block, the per-line test list and the source together
• Markup blocks go through a builtin read loop instead of cat
• 9.5 s → 4.5 s, or 58.7 s → 4.5 s counting perf(coverage): escape the HTML report once per file, not twice per line #1097
• Byte-identical on a fixed corpus, verified both with no hits and with real hits and tooltips — the no-hits corpus never renders a badge, so it alone would not catch a mistake there
• The badge's × is passed as an awk variable rather than \xc3\x97, since \x is not POSIX awk

[Chemaclass]

🤔 Background

Related #1098

After the escaping fix, generate_file_html was 8.1 s of a 9.5 s report: a
Bash loop classifying, looking up and echoing markup for each of 22,405 source
lines, plus 15 cat <<EOF blocks per page — 1935 forks for a 129-page report.

💡 Changes

• The code table is one awk pass, reading the hits block, the per-line test list and the source together
• Markup blocks go through a builtin read loop instead of cat
• 9.5 s → 4.5 s, or 58.7 s → 4.5 s counting perf(coverage): escape the HTML report once per file, not twice per line #1097
• Byte-identical on a fixed corpus, verified both with no hits and with real hits and tooltips — the no-hits corpus never renders a badge, so it alone would not catch a mistake there
• The badge's × is passed as an awk variable rather than \xc3\x97, since \x is not POSIX awk