spike: review fpexif codebase and guage interest

[pgray]

Hello!

I've emailed with @clanmills a bit recently and wanted to reach out to get some feedback on a pure rust rewrite of exiv2 (and exiftool) I've been hacking on in my spare time with claude. I gave a talk on it a bit at SRUG (Seattle Rust User Group) recently.

It's a completely AI-generated experiment right now and the goal was read-only operations but I've got it to some high levels of passing tag matching for exiv2/exiftool. I also relied on macros to reduce boilerplate.

I am going to get exiv2json plugged into my test suite this week.

link to talk

I'm using the rawsamples.ch and raw.pixls.us datasets for validation

The repo is private right now but I'd happily invite you folks for review.

I plan to license it according to how you all and exiftool see fit since I used the two codebases as reference code.

Currently there's a working wasm demo in CI as well

[pgray]

Here are some metrics on tag keys/values:

ExifTool Comparison

Format	Files	Match %	✓ Match	✗ Mismatch	⚠ Missing	+ Extra
3FR	3	95.1%	117	0	6	3
ARW	31	73.8%	4827	27	1689	387
CR2	54	77.7%	11468	95	3189	946
CRW	18	15.6%	378	9	2035	85
DCR	1	8.9%	42	2	427	13
DNG	13	54.0%	1026	30	845	156
ERF	1	80.7%	46	0	11	0
KDC	5	79.9%	226	4	53	77
MDC	0	N/A%	0	0	0	0
MEF	1	87.8%	43	6	0	0
MOS	2	10.2%	30	0	263	12
MRW	8	60.4%	530	13	335	71
NEF	47	83.9%	7605	17	1438	776
NRW	4	90.1%	525	6	52	45
ORF	36	82.9%	5578	3	1150	128
PEF	17	37.3%	1254	20	2089	199
RAF	30	90.3%	3241	0	348	161
RAW	11	83.2%	787	72	87	261
RW2	20	81.8%	3062	200	483	110
SR2	1	75.3%	73	1	23	4
SRF	1	56.8%	42	1	31	0
SRW	7	47.1%	339	21	359	16
X3F	8	46.5%	353	7	399	11

exiv2 Comparison

Format	Files	Match %	✓ Match	✗ Mismatch	⚠ Missing	+ Extra
3FR	3	32.0%	47	52	48	18
ARW	31	27.7%	1349	338	3179	3466
CR2	54	22.9%	2396	3541	4509	6062
CRW	18	1.3%	18	36	1346	224
DCR	1	20.2%	17	15	52	6
DNG	13	29.2%	432	212	836	3239
ERF	1	37.3%	22	18	19	5
JPG	1	52.2%	12	4	7	2
KDC	5	46.3%	126	58	88	38
MEF	1	29.5%	26	11	51	7
MOS	2	34.9%	22	2	39	2564
MRW	8	41.6%	288	85	319	246
NEF	47	18.5%	1687	3424	3990	3079
NRW	4	30.1%	145	126	211	297
ORF	36	39.3%	2275	1980	1539	1145
PEF	17	37.1%	781	431	891	244
RAF	30	41.5%	1262	724	1053	541
RAW	11	28.2%	136	73	274	303
RW2	20	21.7%	521	669	1216	1085
SR2	1	36.7%	33	7	50	36
SRF	1	66.7%	30	5	10	3
SRW	7	22.9%	151	30	479	145
X3F	0	N/A%	0	0	0	0

[clanmills]

@pgray Patrick: Can you add some explanation to those tables? I've understood the format/file columns. What are Match, Mismatch, Missing and Extra? I'm interested to see how the tvisitor.cpp code in my book stacks up against exiv2 and exiftool.

As you know, from our correspondence, I am interested in re-writing tvisitor.cpp in Rust. At present, tvisitor reports metadata and has no code to modify images. Rewriting metadata blocks in a robust/reliable manner is a considerable undertaking which Andreas faced with skill and total success. tvisitor.cpp was left read-only for a couple of reasons. Firstly, the book "Image Metadata and Exiv2 Architecture" is focused on file and metadata standards, so modifying metadata is beyond scope. Secondly, although modifying metadata is discussed in the book, I chose to save months of work required to write and test the code to modify the metadata.

It would be wonderful to see Exiv2 live into the future as the "go to" solution for reading/modifying image metadata. I will be happy to contribute to such a project.

Alison and I are both in good health and excellent spirits. I celebrated my 75th birthday on 18 January and Alison will reach that milestone in July.

[pgray]

Hey Robin,

The "Match" is for tags (e.g. "ISO) that match for fpexif/exiftool/exiv2. The "Mismatch" have mismatched tag values. The "Missing" are tags that exist in exiftool/exiv2 output. The "Extra" are tags that exist in fpexif output but not exiftool/exiv2.

So basically I've got a codebase that can read a ton of different file formats and I've had claude reworking it to properly match both exiftool and exiv2 based on operation mode.

I'm hoping I can get some design input to make the codebase more friendly to read and maintain or otherwise integrate-able.

Right now it's still a bit of a science experiment

[pgray]

Hey folks, here's the codebase: https://github.com/pgray/fpexif

It's somewhat rough but also is functional and all safe with a working wasm demo

I hope to sync up sometime soon

Cheers