Before Uproot 5.3.7, uproot.update corrupted the TFile::fVersion

[jpivarski]

Uproot 5.3.7 adds only PR #1212, which fixes a bug in uproot.update that corrupted the fVersion property of the TFile.

You can check for affected files with

% python -c 'import uproot; print(uproot.open("broken-version-number.root").file.fVersion)'
-937600

or

root [0] auto f = new TFile("broken-version-number.root");
root [1] f->GetVersion()
(int) -937600

This is supposed to be a number like 62400 (for 6.24/00).

This bug will likely affect files larger that 4 GB in Uproot and ROOT, and it has some other subtle effects in ROOT. ROOT checks the file version number for old files that might have old features, and a negative number is below those thresholds. (PR #1212 goes into depth about one such bug.)

Updating to Uproot 5.3.7 will prevent new files from getting a corrupted TFile::fVersion, but not files that were made before the patch. It is possible to cure old files with code like

>>> import uproot
>>> file = uproot.update("broken-version-number.root")
>>> file.file._cascading.fileheader._version
-937600
>>> file.file._cascading.fileheader._version += 1000000
>>> file.file._cascading.fileheader._version
62400
>>> file.file._cascading.fileheader._file_dirty = True
>>> file.file._cascading.fileheader.write(file.file.sink)
>>> file.file.sink.flush()

But be careful when you apply this fix, making backup copies of valuable files before you do so.

If you have an original, broken file named original-broken.root and a copy of it that has been fixed named now-fixed.root, here's how to see that you've only updated the version number and nothing else:

% hexdump -C original-broken.root > original-broken.root.hexdump
% hexdump -C now-fixed.root > now-fixed.root.hexdump
% diff original-broken.root.hexdump now-fixed.root.hexdump
1c1
< 00000000  72 6f 6f 74 ff f1 b1 80  00 00 00 64 00 00 56 38  |root.......d..V8|
---
> 00000000  72 6f 6f 74 00 00 f3 c0  00 00 00 64 00 00 56 38  |root.......d..V8|

Only the 4th through 8th bytes should change. ff f1 b1 80 is -937600 and 00 00 f3 c0 is 62400:

>>> np.uint32(0xff_f1_b1_80).view(np.int32)
-937600
>>> np.uint32(0x00_00_f3_c0).view(np.int32)
62400