Skip to content
Fetching contributors…
Cannot retrieve contributors at this time
155 lines (145 sloc) 11.2 KB
TODO: MDR: Use SPARQL endpoint to query available PUID's, then fetch them (URL:{%0D%0A+%0D%0A+%3Fs+<>+%3Fname+.%0D%0A+%3Fs+<>+%3Fext+.%0D%0A+OPTIONAL+{+%3Fs+<>+%3Fpuid+.+}%0D%0A+OPTIONAL+{+%3Fs+<>+%3Fxpuid+.+}%0D%0A+%0D%0A%0D%0A}&output=xml&jsonp=&key=)
TODO: ANJ: Complete validator to check refs, files, metadata and regexes are well formed, match EOF/BOF/VAR etc.
TODO: ANJ: Report inconsistency of STATUS flag?
TODO: ANJ: When in -checkformats mode, use levels of XSD validation instead of much of the logic in code.
TODO: ANJ: Run DROID, file, and others(?) against a fileset and normalised output for comparison?
TODO: ANJ: Add a format data conversion option to skip the <details> part.
TODO: ANJ: Strip rich format records out for speed - keep data elsewhere.
TODO: ANJ: Get usage of Dublin Core reviewed.
TODO: Handle terminal backslash in a signature
TODO: Change order of preference for extension matches!
TODO: Review api against JHOVE/2
TODO: Review format information against FITS
TODO: Get the magic file and add signatures from it
TODO: Move the fetch code from shell/wget to python
note: In 0.9.3 I tried memoizing the routines to get_puid/regex/etc. This resulted in an overall slowdown.
done: Fix fmt/40 - for word; it requires 'microsoft word', but my example has 'microsoft office word'
note: As of 8-Dec, there are 52 identical regexes.
done: Handle whitespace in xml - what is proper for regex? Is there a good way to preserve it?
done: No extra escaped backslashes in formats.xml
done: Make sure that the fixups get applied; at the moment, they are in formats.xml, but may be lost
skip: Tried, but slowed things down. I noted that some signatures share byte sequences. Check how many there are any if more than a few, handle properly.done: Handle bigendian. x-fmt/392 JPEG2000 is the only one; consider reworking sig and dropping feature
skip: Use int ids, offsets, etc, not strings.
skip: Consider using mmap instead of reading the buffer
skip: Skip zero-len files
skip: Try some alternate implementations of format checking. E.g., no-compiled regex, try-me-else, first n-byte hash, reverse end
skip: Try multiprocessing or threading to do IO and matching in parallel; the 0.5 setup doesn't seem to max out the CPU
skip: Rejig the to implement a single function check_formats that is free to be optimised.
skip: Handle tar, jar, gzip container formats / transforms as in METS
skip: Handle end-tests efficiently. Or review them for other optimisations. Reverse? s[::-1]
skip: Add a no EOF option (perhaps a bad idea until sigs are improved)
skip: Add thread to handle signal and report on current file, #done, #todo, time
done: Add in extension handling / requires modifying prepare as well as run.
done: Produce a single file app / .exe. Well, there is now a script fido.bat and
done: Fix parseReport so that it doesn't create a new parser for each file. Not worth the trouble.
done: Rename _parse to _convert
done: Consider if we should use the non-greedy qualifiers *?, +?, ??, or {m,n}?; consider (?i). Doesn't seem to matter in general.
done: Find out howone to distribute the windows installer though github.
done. Evaluate current droid signatures to see which features are used
done. Parse the sig file, building up an internal representation
done. Match a sig against a string / byte array
done. Move sig xml to conf
done. Handle offset
done. Offsets are 512, 4, 10 - so no long ones; use .{Offset}
done. Match a sig against a file / stream
done. Loop over sigs, find a match, report it
done. Reuse buffer for start/end of small files
done. Save parsed sig file and reload for increased speed
done. Parse sig priority
done. Parse formats with multiple signatures and multiple bytesequences (e.g. pdf/1.7 fmt/276)
done. Replace ^ with /A and $ with /Z
done. Handle one signature at a time, reading in the BOF and EOF into separate buffers.
done. Why doesn't PDF 1.3 EOF match sample. Because of eof v bof match behavior.
done. Handle PositionType=Variable. Variable means the sequence can be anywhere, i.e. .* prefix.
done. Fix EOF matching! 'abc\Z' doesn't match 'xxxabcd', but '^ab' does match 'abxxx' But by starting with .*
done. Sort formats by priority so that the first match is the best one.
done. Add basic timing information. Running on sample directory at 60+ files/sec.
done. Bug. didn't handle [a-b]{m,n] seen on c:\Documents and Settings\afarquha\My Documents\Meetings\2009\2009.05.25 IASSIST\iassist_v4_24b.mp3
done. Do some profiling to see if it is worth doing any optimisation. Not now. Memory usage is very low and stable for 2k files. Speed is around 50 files/sec.
done. Handle zip.
done. Handle multiple hits and removing lower priority items; RTF get 7 format matches; TIFF get 4. Do the OBVIOUS fix and just remove-if. Speed is OK for now.
done. (I think) Check that offset/maxoffset are correctly handled
done. Check the meaning of offset=0 with EOF. Do we have any other than seqID=242
done. droid recognises test.tar as a tarfile x-fmt/265. For pronom [a-z]{1-5} = [a-z].{1,5} !
done. droid recognises buckland as html fmt/99; Incorrect handling of offset=0, maxoffset=n
done. alani-* pdf not recognised. Droid lists all pdf formats due to ext match. Error in signature.
done. v1.docx - fido sees zip, droid sees docx. One signature looks for the ooxml indicator, the other for zip. The zip signature differs from the real zip signature. I think that the BOF should be EOF!
done. ppt - fido sees ole2, droid sees ppt. The sig looks for a pattern that shows up at about 500kb, outside the default buffer. Better to improve the signature.
done. Test fidotests/corpus against droid results until they are the same. They are correct, but not identical.
done: Add more flexible output syntax (match droid?)
done: Add timing to output at file, format, sig level.
done: Cleanup after zip properly. We currently leave empty directories in the tempdir. No longer copy to files, but just read.
done: why are some files (always ppt?) so slow? E.g. *v6B JM.ppt (might just be a bad signature)
When the bufsize goes from 7500 -> 7800, runtime goes from 100ms -> 20315ms!
Some bad non-linearity in the regex package
The offender appears to be FormatID=u'687',FormatName=u'MPEG 1/2 Audio Layer 3'
And other mpeg signatures. They have multiple {36-1426} or so
done. Allow add/del/mod of pronom sigs. Use
done: change to have conf/pronom and conf/local where local adds/changes/deletes the pronom info. Using format_fixup.
done: Let prepare take xml from the zip. This is to simplify distribution.
done: Introduce a lite syntax for signatures using just the needed information. Using
done: Load pronomlite and over-ride or extend the pronom info. Using
done: Update file locations using conf. None to update, I think.
done: Handle large zip file items by buffered-read to end (if fs=75, bl=33, read 33, 9, buf=read(33)
done: Provide a files=- which reads the filenames from stdin
done: Move the xml to a zip file. Would be good, but there isn't a good way to get the items without creating the file.
done: One of the PDFs ends with \n\r, instead of \r\n and the regex fails. What is the right thing to do? Change the regex to 'EOF..?. ?
done: Add way to show formats, select formats, exclude formats.
done: Remove from the distribution; (done: create README.txt); add how-to
done: Correct the and directory configuration so that Fido installs for linux
done: Parse the Droid-v40 and check for any new / updated format definitions.
done: Tidy interface to the (no)print options.
done: Recursive zip handling - extract internal zip to a tempfile and recurse; remove on exit.
done: Handle tar as well as zip
done: Rejig the file processing to us an iterator that could then read from a queue, stream, etc.
done: Add in some unittests
- How can I get the string that r or b prefixes? These are much more readable than repr(s).
- How can I pickle the re.compile() results and the file
- It looks as if there is only one pronom 'is subtype of' statement!
- One of the .mov files has ftypqt which is probably quicktime. There is not signature for this! There is an mp4a at offset 591, 599.
- The indirectoffset is not used
- At least some of the variable patterns really seem to have bof+maxoffset nature
- fmt/50 and fmt/51 (RTF) have identical signatures and no precedence.
- zip x-fmt/263 and JAR x-fmt have no precedence relation. But jar > zip
- Check JP2 in c:\Documents and Settings\afarquha\My Documents\DPT\Image Comparison\000833138_01_000021.jp2
- Add signatures for: .txt, .sh, .css, .bib, .tex, .java, .py, .c, .dot (msft), .*~, .dot (graphs), Makefile, .bz2, .aspx, .js, .warc, .wgz, .xsd, .ini, .mm, .tab,
- Add signatures for: .azw .mbp (kindle), .itdb .itc2 (itunes), .ost, .bat, .sql, .eps, .sam, .ico
- Review sigs for: .pdf (several fail), .htm, ~*.doc, .rtf, .mp3, .au, ~$....docx, .ps,
- The sig for tar seems poor. There are not separate sigs for the major tar formats (ustar, gnu, pax); it doesn't recognise my sample.
- There isn't a sig for a tar.gz file, so it is just recognised as a gzip.
- The docx signature has a BOF .* which should be a EOF, as it is in zip. If the BOF buffer is big enough, then it won't matter.
- The ppt signature has a variable P.o.w.e.r. .P.o.i.n.t pattern that shows up in file rather late, after the buffer ends. Fido is OK. Better to improve the sig.
- Review all of the Microsoft format signatures
- The doc signatures could have a EOF 'Microsoft Office Word Document\x00.*\Z' or similar
- Signatures needed for specific types of OOXML such as docx, pptx, xslx, dotx, and the macro versions thereof
- regex (?iLmsux) - (One or more letters from the set 'i', 'L', 'm', 's', 'u', 'x'.)
The group matches the empty string; the letters set the corresponding flags:
re.I (ignore case), re.L (locale dependent), re.M (multi-line), re.S (dot matches all),
re.U (Unicode dependent), and re.X (verbose)
- Droid produces a csv with columns:
- zip items have a uri of zip:file:/file path!/zip path (but no File_PATH
- The CONTAINER formats such as zip, jar, gzip are not specified as such in pronom, but only in the code.
BOF -> \A - better than ^
EOF-> \Z - better than $, which messes up with multiline
?? -> .?
* -> .*
{i} -> .{i}
{i-j} -> .{i,j}
{i-*} -> .{i,}
[aa:bb] -> [aa-bb]
(abc|abd) -> (?:abc|abd)
The following are define in the language, but never used
[abc:abc] -> - There are no [bbb:
[!a] -> [^a] - There are none
[!abc] -> (?!abc) - There are none
[!a:b] -> - There are none
Indirect offset
Jump to Line
Something went wrong with that request. Please try again.