When a file transfer is resumed, as a first step a HEAD request to the patch endpoint is made to determine where to resume.
In the previous implementetation the offset was not calculated in the right way. That way the resume offset was lower than necessary.
The problem was, that the glob result is sorted lexicographically. As a consequence some offsets were determined as not present, because the previous one comes in a later iteration.
Here is a small example:
glob result (folder path stripped): [.patch.0,.patch.10000000,.patch.15000000,.patch.5000000]
Before change:
$last_offset after loop: 5000000 (loop breaks in second iteration)
$offsets after loop: [0,10000000]
After change:
$offsets after sort: [0,5000000,10000000,15000000]$last_offset at the end: 20000000
When a file transfer is resumed, as a first step a HEAD request to the patch endpoint is made to determine where to resume. In the previous implementetation the offset was not calculated in the right way. That way the resume offset was lower than necessary.
The problem was, that the glob result is sorted lexicographically. As a consequence some offsets were determined as not present, because the previous one comes in a later iteration.
Here is a small example: glob result (folder path stripped):
[.patch.0,.patch.10000000,.patch.15000000,.patch.5000000]Before change:
$last_offsetafter loop:5000000(loop breaks in second iteration)$offsetsafter loop:[0,10000000]After change:
$offsetsafter sort:[0,5000000,10000000,15000000]$last_offsetat the end:20000000