GitLab

Update dependency mocha to v9.2.2

See merge request !26

Update dependency mocha to v9.2.1

See merge request !25

Update dependency mocha to v9.2.0

See merge request !24

Update dependency mocha to v9.1.4

See merge request !23

Update dependency mocha to v9.1.3

See merge request !22

Update dependency mocha to v9.1.2

See merge request !21

Update dependency mocha to v9

See merge request !20

Re-enable ifps-deploy logs for review app.

See merge request !19

https://github.com/ipfs-shipyard/ipfs-deploy/issues/215

Previously average quality was divided by the size. This means that subdividing a grid could result in an equivalent score even if the newly added cells were all zero! This is because the average would be half, but it would be divided by half less due to the small grid bonus. The only thing really protecting us was that we needed some good lines to "propose" a grid that small.

This was fixed by adjusting the bonus to scale slower. The current factor was picked so that when subdividing the grid the newly considered points need to be at least half as good as the previous points. This factor sounds a bit low but it makes sense as we don't want to get hung up on some really bold lines that are part of the style instead of the grid. In general we expect the grid to be fairly consistent so we don't want to give too much preference to a small number of bold points.

Fixing this factor also allowed us to consider more candidate distances. Previously lines slightly off of the real grid could trigger us considering a very small grid, over-ranking it and selecting it. Now that we have fixed that considering small grids is not very harmful. The max number of lines to consider was chosen to be slightly less than the square root of the total lines. This effectively counteracts the fact that we consider every pair of chosen lines which is `O(n^2)` resulting in the overall algorithm being less than linear with the image dimensions.

This has two main components.

We de-anti-alias the line by shifting the pixel location from the middle to the edge.

Secondly we try to find the largest set of lines for a given width to base the distance on. If we find two candidates far apart we can get a really good guess for line size.

This works much better. The biggest issue is that it doesn't support non-integer grid sizes. I think that we can support that in the future as well but "de-aliasing" the selected edges. By finding the "center" of the line I think we can estimate non-integer sizes fairly well.

Also adds some debugging views.

- Overall works better.
- Works for light gridlines.

This only really works for local files. Ideally we would set the `crossorigin` attribute on the `<img>` but that seems to cause requests to fail even though https://developer.mozilla.org/en-US/docs/Web/HTML/Element/img#attr-crossorigin seems to say that it should be no worse. So until I figure out what I am doing wrong I am not setting the attribe which basically prevents all URLs from auto-detecting.

Fixes #3.

I am not a UX designer.

Update dependency mocha to v8.4.0

See merge request !18