WIP: Fix #1329: Update taffy to 0.5.2

[iamralpht]

Taffy changed the way the "measure function" is passed in, and it actually now matches how the JNI glue was doing it already.

[iamralpht]

I think the Battleship sample might be rendering differently, so pushing to confirm/deny that via the CI tests.

[github-actions]

Snapshot diff report vs base branch: main
Last updated: Sun Jul 14 23:01:03 PDT 2024, Sha: a1fc390

File name	Image
Grid-Layout_compare. png
Fill-Container_compa re.png
Dials-Gauges_compare .png
Battleship_compare.p ng
ChangeFileScreen_com pare.png
ExpandedSwitcher_com pare.png

[timothyfroehlich]

Happy to see the update to the current version, but what shall we do about the layout changes? Most of them look fine, but obviously the Battleship one is odd.

[timothyfroehlich]

Happy to see the update to the current version, but what shall we do about the layout changes? Most of them look fine, but obviously the Battleship one is odd.

Oh, just noticed that you'd labelled it "WIP". Mind if you switch the PR to Draft? I assumed you wanted reviews right now.

[iamralpht]

Oops — yes, I requested reviews too soon and then changed to WIP once I saw more failures than I could fix last night… I will move it to Draft now.

…

On Mon, Jul 8, 2024 at 07:27 Tim Froehlich ***@***.***> wrote: Happy to see the update to the current version, but what shall we do about the layout changes? Most of them look fine, but obviously the Battleship one is odd. Oh, just noticed that you'd labelled it "WIP". Mind if you switch the PR to Draft? I assumed you wanted reviews right now. — Reply to this email directly, view it on GitHub <#1330 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAQYTY5XFGNHL3OSTEY6F5TZLKOUZAVCNFSM6AAAAABKQBI5AOVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDEMJUGIZTAMBWGM> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[iamralpht]

It took me quite a long time to figure out where the changes were coming from. There were two sources:

• Our text measurement function would sometimes return a large height (depending on the incoming Available Size), which the text node itself would forget, but which would expand the text node's containing frame in some cases. I addressed this with a more careful interpretation of Available Size in the DesignText measure function. I was able to find this by tracing the measure function and changing its return values (before fixing the Available Size interpretation).
• If a flex node needs to shrink in the cross axis (e.g.: the big battleship frames) to be smaller than its children, then we need to set width or height (as appropriate) to something other than Auto. Zero works. I eventually found this issue after making a tool to dump a layout tree into HTML and comparing our layout with Chrome's. Chrome generates the correct layout with auto, but also generates the correct layout with zero.