Overlap 2 lines on full page scroll (page up/page down)

[amakrushin]

Steps to reproduce:

1. Open a diff, which not fits one screen
2. press Page Down

Expected: 2 lines overlap from the last screen
Actual: full page scroll, no overlap.

Note: not all editors overlap:

• emacs, vim - overlap
• less - no overlap.

emacs:
To read the buffer a windowful at a time, use C-v with no argument. It takes the last two lines at the bottom of the window and puts them at the top, followed by nearly a whole windowful of lines not previously visible. If point was in the text scrolled off the top, it moves to the new top of the window. M-v with no argument moves backward with overlap similarly. The number of lines of overlap across a C-v or M-v is controlled by the variable next-screen-context-lines; by default, it is 2.

[umputun]

the behavior you describe is real, pgdown scrolls by exactly the viewport height with no carryover.

before I take it though, I need a case that isn't editor parity. The emacs/vim citation doesn't move me on its own, and you note yourself that less doesn't overlap, so the tools disagree with each other here.

the reason emacs overlaps is that C-v destroys your anchor, point moves to the new top of the window, so two lines get carried over to rebuild it.

revdiff doesn't lose that anchor. pgdown scrolls by the cursor's own visual delta, so the highlighted cursor line stays on the same screen row across the page. That was a deliberate fix for #124, and it's why the emacs comparison doesn't map cleanly.

so what I'm after is the concrete friction. Where did the missing seam actually cost you something? Reading a hunk that straddles a page edge and losing the thread would be a real case, and that would change my read.

related: #278 asks for a configurable scroll-off, same family of setting. If this goes in it'll be shaped together with that one, and off by default.

[amakrushin]

After the full page scroll, my eyes automatically search the overlapped lines at the top.

Without overlap, I am not sure, maybe I double clicked and scrolled extra page.

[amakrushin]

btw, In less, pgup, pgdown scroll half page, so I didn't notice that less do not overlaps until today

[umputun]

that second note is the answer, I think.

half page in less is d / ^D, and revdiff binds ctrl+d and ctrl+u to the same motion, on by default, no flag: the viewport moves half a screen and the bottom half of what you were reading becomes the top half of the next screen. On a 24-row terminal that is about 11 lines of carryover rather than 2.

so the seam you look for at the top is already there, on the key you are already pressing in less. Ctrl+d/Ctrl+u is in the README key table and in the ? help overlay.

that also covers the double-press worry: half-page scroll makes an accidental extra press obvious, because a doubled ctrl+d still leaves you a full screen of context you have already read.

if what you want is specifically a full page plus a two-line seam, that is a different request from the one your less habit describes, and worth saying so explicitly. Otherwise I would rather not add a config knob for something already bound.

[umputun]

changed my mind on this one. The same ask came in from other people since, both in the repo and over direct channels, so the demand is real enough to add the knob.

implemented as --page-overlap N in #313, merged to master. It subtracts N from the full-page motion, so PgUp/PgDn carry the bottom N rows of the screen to the top of the next one. Default is 0, so paging is unchanged unless you set it. Also available as REVDIFF_PAGE_OVERLAP and as page-overlap in the config file.

two things worth knowing:

• ctrl+d / ctrl+u are excluded. They already keep half a screen, so subtracting an overlap on top would shrink an already-overlapping motion.
• the carryover is approximate on wrapped or annotated lines. The walk stops on cursor positions and one position can span several rendered rows, so a tall block at the page edge carries over more than N when the walk rolls back off it, and less than N when it takes the block whole. In a plain unwrapped diff it is exactly N.

not in a tagged release yet, it will be in the next one.