Fix dagre.js for abnormal wide edges

[seanshpark]

This will update dagre.js for abnormal wide edges

[seanshpark]

this change has suggestion from @beru at #1246 (comment)

but still this part goes bad

before	after

• from the model in Fix dagre.js for abnormal wide edges #1246 (comment)

[seanshpark]

I'm still reading the papers of Dagre ... and looks like will take a while.

[beru]

@seanshpark

The code I wrote in #1246 (comment) didn't make much sense so I did some experiment. https://github.com/beru/netron/commits/horizontalCompaction/

It was difficult for me to understand buildBlockGraph and the second pass of horizontalCompaction so I decided to write another routine to remove unused space. The code quality is low but don't throw stones at me.

[seanshpark]

@beru, Wow, you've got the sample working !

It was difficult for me to understand buildBlockGraph and the second pass of horizontalCompaction

Algorithms are described in the papers in https://github.com/dagrejs/dagre/wiki#recommended-reading (I think you may know the links) and still reading these hard-to-understand algorithms in my spare time ...
I'll also take a look at your code :)
BTW, what about posting a PR ?

[beru]

Algorithms are described in the papers in https://github.com/dagrejs/dagre/wiki#recommended-reading (I think you may know the links)

Thank you for the info. I didn't know the wiki of dagre project or maybe I just forgot it? Anyway, I'll try to understand what's written on the paper.

BTW, what about posting a PR ?

I wish to do so if I'll be able to come up with more polished code. The code I wrote doesn't use the second pass at all (as I'm not capable of understanding it) but that's probablly not a good idea.

[seanshpark]

@lutzroeder , I've update the patch with what I understand as of now.
For the problem of undesired long edges are from the dummy nodes in dagre.
This change will skip for calculating the position for dummy elements,
so that they don't go to some unwanted position.

But still there exist a problem with the model you've provided and the result is same as in #1254 (comment) capture.
What I know is that this particular node input/output is reversed in Y position.
I hope I could find the cause to fix the degradation someday :)

If possible can you please land current change?

[seanshpark]

@lutzroeder , I've updated for CI build fail.