[BUG] [REGRESSION]: `tickprefix`/`ticksuffix` ignored when `tickvals` provided

[emilykl]

Codepen: https://codepen.io/emilykl-code/pen/jEBbzvL

In plotly.js v2.29.1 and earlier, tickprefix and ticksuffix would be applied as expected when tick values were provided with tickvals:

However starting with v2.30.0 and continuing to the latest v4.0.0, suddenly tickprefix and ticksuffix are ignored when tickvals is passed:

There's no reason why tickprefix/ticksuffix can't coexist with tickvals, so this should be fixed.

This bug was also noticed in plotly.py issues plotly/plotly.py#4041, plotly/plotly.py#4678

[LeonxLJX]

I'd like to take this one. I just shipped the matching JS-side fix for marker.colorbar.thicknessmode in #8015, so I'm neck-deep in the colorbar/axis tick rendering path already — this is a closely related regression in the same module.

Root cause (best read from the codepen + the issue). Somewhere between 2.29.1 and 2.30.0, the tick label render path stopped applying tickprefix/ticksuffix when tickvals is user-supplied. The user-supplied tickvals path almost certainly short-circuits the autoTickFormatter (which is where tickprefix/ticksuffix get spliced in), and then renders tickvals.map(v => String(v)) directly without re-attaching the prefix/suffix. The fact that it works when tickvals is absent (auto-tick path) and breaks only when tickvals is present is the smoking gun — it's a branch in the tick renderer that forgot to apply the same prefix/suffix transform that the auto-tick branch does.

Alternatives I considered:

• (a) In the tickvals branch, run each value through the same prefix/suffix wrap that auto-ticks get — formatted = tickprefix + fmt(v) + ticksuffix before passing to the SVG text drawer. Smallest patch. Best fix IMO.
• (b) Refactor so there's a single formatTickLabel(value, axis) helper that both branches call. Larger refactor but eliminates this whole class of "branch forgot to apply X" bug; would also fix any sibling regressions lurking in the auto-tick path.
• (c) Restore the 2.29.1 path verbatim and add a regression test — useful as a bisect anchor but doesn't tell us why it changed, so the next refactor reintroduces the same bug.

Suggestion to maintainer. I'd go (a) for the immediate regression (matches the spirit of the existing auto-tick code) and open a follow-up issue to evaluate (b). The fix is ~5 lines in src/components/axis/, plus one new test in tests/image/mocks/axis_tickformat_*.json that exercises tickvals + tickprefix + ticksuffix for a linear and a date axis. I have the test fixture scaffolding from #8015 so I can adapt the same approach here. Two things worth confirming before I open the PR: (1) does the date-axis formatter path also need the same treatment, or is that branch already correct? (2) For (b), would you prefer the helper sit in axis/format.js or a new axis/tick_label.js to keep diffs small?

(claiming via @LeonxLJX)