Skip to content

Write a string in whichever form is shorter - #6

Merged
tannevaled merged 1 commit into
mainfrom
string-encoding
Aug 25, 2026
Merged

Write a string in whichever form is shorter#6
tannevaled merged 1 commit into
mainfrom
string-encoding

Conversation

@tannevaled

Copy link
Copy Markdown
Contributor

A PDF string may be written between parentheses or in hex, and the writer only
knew the first. A literal string spells every unprintable byte as a
four-character octal escape, and text that is not Latin-1 is written in UTF-16
— so a string of ordinary prose in any other script came out four times its
own length. The hex form costs two characters a byte whatever the byte is, so
the writer now measures both and takes the shorter.

The corpus makes the case: one file carries a diagram's whole source in its
information dictionary, and rewriting it produced 3.7 MB from 1.2 MB, almost
all of it octal escapes. Across the whole corpus the rewrite drops from 98.9%
of the input to 98.8%, which is what a change that only bites on non-Latin
text looks like in an English-language corpus; on the file that motivated it,
the string itself halves.

All 118 833 files still rewrite to an identical fingerprint. Exact 100%
statement coverage, go vet clean.

A PDF string may be written between parentheses or in hex, and the writer only
knew the first. A literal string spells every unprintable byte as a
four-character octal escape, and text that is not Latin-1 is written in UTF-16
— so a string of ordinary prose in any other script came out four times its
own length. The hex form costs two characters a byte whatever the byte is, so
the writer now measures both and takes the shorter.

The corpus makes the case: one file carries a diagram's whole source in its
information dictionary, and rewriting it produced 3.7 MB from 1.2 MB, almost
all of it octal escapes. Across the whole corpus the rewrite drops from 98.9%
of the input to 98.8%, which is what a change that only bites on non-Latin
text looks like in an English-language corpus; on the file that motivated it,
the string itself halves.

All 118 833 files still rewrite to an identical fingerprint. Exact 100%
statement coverage, go vet clean.

Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
@tannevaled
tannevaled merged commit 5d74c02 into main Aug 25, 2026
1 check passed
@tannevaled
tannevaled deleted the string-encoding branch August 25, 2026 11:02
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant