tuetschek / en-deep Goto Github PK

The line numbers doesn't even reflect the actual position (i.e. the
currently parsed task).

Reduce the usage of obsolete synchronized collections.

There should be a "legal" way to stop the running workers, such as special file 
that the task should check for existence upon each task retrieval.

StToArff should clear all the output files before each run.

Task reset feature should be reviewed, its behavior is sometimes strange (for a 
prefix, it resets the whole plan).

If there is a change in the scenario file, it should be recognized upon
task reset (just for the tasks that are to be reset).

Issue warnings if there may be a pattern collision (one pattern is a
subpattern of another) in order to prevent problems in scenario reruns.

E.g.: file*.txt and file1*.txt exist in two different tasks, but file1*.txt
is produced later. If file1*.txt is already produced and the task that
produces file*.txt gets reset, the pattern expansion includes also
file1*.txt and it may get messy.

There should be some switch with path prefix that will be considered to be 
local, e.g. /tmp. All tasks which have this prefix in their I/O specs and 
depend on each other should then be lined up for computation on the same 
machine.

In the current setting, the expanded task is copied along with all the 
dependencies of the original task, which are then removed. This increases 
complexity and reduces performance with more than ca. 10000's of tasks.

The program is yet unable to handle UTF-8 characters in the input ST files.

If an input like this is provided:

params: lang_conf="st-en.conf", omit_semclass="1", predicted="1",
pred_only="1" generate="Children";

then the parser doesn't report an error, but the last parameter is not
recognized at all.

The parser should report an error.

GreedyAttributeSearch does not work well together with attribute rankings that 
do not contain all attributes.

Add new generated feature: children patterns without function words.

There should be an option, which, when selected, just parses the scenario file 
and end the whole program. It would be useful just to check for errors in the 
plan file before launching the process.

For a task that outputs a-**.txt, some of which are a-*-x.txt, there is no way 
to capture just the a-*-x.txt in the input of another task. 

There should be something like a-*|-x|.txt, which would depend on a task 
producing a-**.txt, but take only a-*-x.txt as input

In the current version, sub-specifications of prefixes and suffixes for the 
expansions do not work (only if they're at the beginning of the expansion 
transitive line).