by the numbers

By the numbers

By Artem.

Snapshot: 2026-10-17, d57306903d7a122f3abe84c4d5d9659799d6176f (0.3.0). These are repository measurements, not a performance promise or an individual contributor ranking.

Scope and reproducibility

Inventory comes from git ls-tree -rz d573069; contents are the corresponding Git blobs, not build products or this wiki's uncommitted refresh. A binary is a blob containing NUL in its first 8,192 bytes: exactly fonts/DejaVuSans.ttf, tests/corpus/truncated-hmtx.ttf and tests/corpus/truncated-cmap-184.ttf.

All size/language tables exclude all droid-wiki/ and docs/evidence/, including evidence READMEs as well as generated logs. They retain ordinary documentation, the ADR, fixtures, examples, research harnesses, licenses and build files. Lines mean newline bytes (wc -l semantics), including comments/blanks; 13 XML fixtures lack a final newline. Bytes are uncompressed blob bytes. These are not logical source-line counts.

Measure	Value
All tracked files	811
Excluded wiki / evidence files	41 / 646
Binary assets excluded	3
Measured text files / lines / bytes	121 / 25,304 / 1,110,939
Production C and headers, src/ + include/	19 files / 6,437 lines
Test/research code, tests/	45 files / 16,147 lines
Build/configuration files, defined below	15 files / 844 lines
Code across production, tests and examples	22,980 lines
Largest measured text file	tests/test_stream.c, 3,655 lines

Production is nine C translation units (eight library units plus src/main.c) and ten headers. Test code is 15 C files (9,646 lines), one C++ file (157), 21 Python files (5,076) and eight shell files (1,268); this includes research tools, not just shipped-feature regression tests. Their ratio is 2.51 test/research lines per production line, not a coverage percentage.

Build/config means .github/'s five YAML files, CMakeLists.txt, tests/consumer/CMakeLists.txt, all four cmake/ templates, Makefile, .gitignore, baseline-manifest.json and libki_xml_pdf.map. Categories overlap: the consumer CMake file is also under tests/. “Code” means .c, .h, .cpp, .py, .sh only.

Lines by language

xychart-beta horizontal
    title "Tracked text lines, wiki and evidence excluded"
    x-axis ["C", "Python", "Markdown", "Shell", "C headers", "YAML", "CMake", "Other", "C++", "Make", "JSON", "XML"]
    y-axis "Lines" 0 --> 16000
    bar [15702, 5076, 1319, 1268, 761, 358, 215, 184, 173, 118, 97, 33]

Loading

Language/type	Files	Lines
C	28	15,702
Python	21	5,076
Markdown	14	1,319
Shell	8	1,268
C headers	11	761
YAML	5	358
CMake	5	215
Other	8	184
C++	2	173
Make	1	118
JSON	1	97
XML	17	33
Total	121	25,304

Classification uses suffixes; CMakeLists.txt and .cmake.in are CMake. “Other” includes cmake/ki_xml_pdf.pc.in, linker map, notices, checksums and expected text. Tiny XML line counts can hide long single-line fixtures such as tests/fixture.xml.

Average file size by directory

Directories are recursive first-level groups; (root) means files directly at the repository root. Same text-only exclusions as above; means are rounded to one decimal.

Directory	Files	Lines	Mean lines	Mean bytes
src/	18	6,201	344.5	10,316.6
include/	1	236	236.0	10,005.0
tests/	61	16,190	265.4	11,674.3
examples/	11	438	39.8	1,157.7
docs/, excluding evidence	7	491	70.1	7,553.3
cmake/	4	59	14.8	570.2
.github/	6	368	61.3	2,160.2
fonts/, notice only	1	78	78.0	3,859.0
(root)	12	1,243	103.6	9,866.1

The largest production files are src/xml_parse.c (1,482 lines), src/pdf_write.c (1,112), src/convert.c (994) and src/font.c (662).

Recent history and churn

All 68 reachable commits are in October: 54 non-merges and 14 merges, on four active UTC dates (October 7: 2; October 13: 8; October 16: 32; October 17: 26). Dates use committer timestamps converted to UTC; activity includes merges.

Recent Monday–Sunday week	Commits
2026-09-07–09-13	0
2026-09-14–09-20	0
2026-09-21–09-27	0
2026-09-28–10-04	0
2026-10-05–10-11 (ISO 41)	2
2026-10-12–10-18 (ISO 42, partial)	66

Recent month	Commits
2026-07	0
2026-08	0
2026-09	0
2026-10, through snapshot	68

Last-90-day churn is git log d573069 --no-merges --since=2026-07-20T00:00:00Z --until=2026-10-18T00:00:00Z --numstat --no-renames. Exclude wiki, evidence and binary entries; retain deleted source paths. It totals 27,083 additions / 1,766 deletions. The initial public import is included, so high churn is not automatically maintenance trouble. Git numstat counts an unterminated final line, unlike the size tables.

Hotspot, ordered by additions + deletions	Touching non-merges	Added	Removed
tests/test_stream.c	19	3,753	98
src/xml_parse.c	4	1,725	243
src/pdf_write.c	12	1,380	268
src/convert.c	13	1,097	103
src/font.c	6	789	127
tests/research_table.py	1	762	0
tests/test_xml_subset.c	1	751	0
tests/test_layout_wrap.c	1	686	0

Explicit bot attribution: a lower bound

Search author name/email and Co-authored-by: trailers, case-insensitively, for [bot], the whole word bot, or factory-droid. There are 0 bot-authored commits, 30 with a bot co-author, and 30 in their union: 30/68 = 44.1% of all reachable commits, or 30/54 = 55.6% of non-merges. The denominator includes the initial import; merges have no matching trailer. All 30 matching commits explicitly credit factory-droid[bot].

This measures visible attribution only. It cannot establish how much code was AI-generated, and absent trailers do not prove work was unaided.

Include depth and export surface

Resolve literal quoted includes relative to their source file, then the project roots include/, src/, examples/; traverse existing tracked C/C++ files and headers only. Exclude system headers, generated build dependencies and evidence snapshots. There are no unresolved quoted includes. Maximum simple-chain depth is four include edges / five files, with six tied paths:

• Each of src/convert.c, src/layout.c and tests/test_stream.c reaches src/layout.h → src/pdf_write.h → src/geometry.h → src/font.h.
• The same three origins also reach src/layout.h → src/pdf_write.h → src/info.h → include/ki_xml_pdf.h.

libki_xml_pdf.map declares 13 function exports, excluding the three ELF version-node names and local implementation/test hooks:

Version node	Functions
KI_0_1 (8)	ki_convert, ki_convert_file, ki_doc_text, ki_doc_break, ki_doc_fail, ki_options_init, xml2pdf_convert, xml2pdf_convert_cap
KI_0_2 (1)	ki_convert_to_file
KI_0_3 (4)	ki_config_init, ki_convert_cfg, ki_convert_file_cfg, ki_convert_to_file_cfg

These are the source-controlled export rules, not nm counts from an arbitrary local build. See C API.

Measurements are revision-scoped

docs/evidence/osr-82-heap-profile/README.md measured de8ba0e, before the October 17 memory changes: median RSS 2,728 KiB for full 16 MiB report-v1, 2,688 KiB for records-v1, and 2,728 KiB for report-v1 limited to 1,000 records. These are not measurements of this HEAD. tests/verify.sh now gates converter RSS at 4,096 KiB, below half the large input and within small-run RSS +4 MiB; that fixture gate is not a universal RSS bound.

Key source files and related pages

Sources	Purpose
include/ki_xml_pdf.h, libki_xml_pdf.map	Version, public layouts and export nodes
src/convert.c, src/layout.h, src/pdf_write.h, src/geometry.h, src/info.h, src/font.h	Production inventory and include-chain evidence
tests/test_stream.c, tests/verify.sh, tests/resource_baseline.py	Test size and measured resource gates
docs/evidence/osr-82-heap-profile/README.md	Historical heap/RSS provenance, excluded from code counts

Overview · Architecture · PDF writer · Resource limits · Testing · Design decisions · History