Robert Sachunsky

GPU processors: (uniformly) log CPU vs GPU use

3

Due to the lack of a specification on that aspect, our processors have no or no uniform way to inform the user whether or not a GPU device is used...

parameter_override_option: handle numeric strings

2

Suppose I want to have: ```json "textequiv-index": { "type": "string", "enum": ["first", "last", "0", "1", "2", "3", "4", "5", "6", "7", "8", "9"], "default": "first", "description": "Only extract lines with...

bashlib: fill ocrd.process.profile logger

1

We are currently blind in our process/profiling logger to any bashlib-based processors (which step ran and how much time was spent). Implementing this with bash is probably not so easy...

processor CLI parser: improve error message for --parameter

2

The ambiguity of `--parameter` (JSON string or filename) currently makes the behaviour very confusing if you pass a filename that does not exist (but is clearly not a JSON string...

parameter decorator: try to resolve path names as resources

1

In preparation of fixing #930. Merely adds new specific test that fails for now. Draft for easier discussion...

bashlib: provide a Workspace.download_file analogue

1

All Pythonic processors use `Workspace.download_file` for their input files for resolving/retrieval. This allows remote URLs to be downloaded ad-hoc to the workspace (without changing the METS reference), but also avoids...

We have `bulk-add` and we have `merge`, but we could really use a `bulk-merge` command: Merge multiple METS files into a single METS at once (without de/serialization in between), perhaps...

parameter decorator: handle preset files as resources

1

When we drafted the resource handling, one of the use cases was installing preset files, i.e. preconfigured (if incomplete) json parameter sets that have a discernable name and can be...

conflicting namespace prefixes during ListRecords

2

If you do a harvest during which the same prefix will be seen with different URL targets, metha-sync will jumble the prefix – suffixing it by `1` but never declaring...

Currently we build minimum, medium, maximum and maximum-cuda in parallel. Except for the latter, this offers no time advantage, just consumes more resources (credits). I suggest we change to -...