Video wird geladen...
Video konnte nicht geladen werden
Agents write code faster than we can review it. A huge PR is essentially a long list of files in alphabetical order. We think reviews can be more fun if you swap the limiting pages for a canvas, which is why we built PR canvas. You read the change... show more
11,688 Aufrufe • vor 6 Tagen •via X (Twitter)
22 Kommentare
@FourEyedWiz @prcanvasreview Really solid!
Join the waitlist:
@prcanvasreview Such a solid product
@prcanvasreview alphabetical order prs are pain fr
reading in the order the commits/logic runs is the right approach. hit the same wall yesterday refactoring a big data model: had an agent dump iterations of the ER diagram to Mermaid just so I could see it, but I couldn’t cause it was huge, so i had to add zoom to my MarkdownPreview to see it. does PR canvas have visualization for data model?
@prcanvasreview Isn’t this what coderabbit does ? I like the canvas thing but coderabbit runs on the repo right there as an extension. 🤷🏽♂️
show unchanged callers alongside changed nodes to help reviewers catch interface regressions.
@prcanvasreview Reading in execution order is the right default for agent PRs. What else helps: have the agent attach the test or log line that proves each hunk, so review becomes checking evidence instead of re-deriving intent.
@prcanvasreview execution order instead of alphabetical. wild that this took so long to exist
@prcanvasreview Reading changes in execution order is a strong idea, but a canvas won’t make review fun if summaries and suggested issues distract from the actual diff. Keep the code one click away.
an audit of my whole repo in august turned up a fix that had gone into two of the three scripts writing my index files and never into the third. it never showed in that pull request, because the third file wasn't in the diff. my rule since then, for a fix like that one, is to grep for every writer and mark each one as fixed, still needing the fix, or exempt. i wrote up a second case, a path fixed in two files and missed in a third, here could the canvas list the other code that writes the same data, even when the pr never touched it?
@prcanvasreview Reading in the order the logic runs is the right call. Alphabetical file lists were never built for humans. Would love a visual diff card for UI changes too, so designers can review agent PRs.
@prcanvasreview Reading in execution order fixes the real cost of agent PRs: an alphabetical diff makes you rebuild the call graph first. The card summary is where I'd be careful. If the agent that wrote the code also writes it, reviewers start checking its explanation instead of the code.
Reading in the order the logic runs beats alphabetical files by a mile. A habit that pairs well with it: have the agent open each PR by naming the two or three spots where a wrong assumption would hurt most, like a migration, an auth check, or a retry loop. Review those first, then skim the rest.
@prcanvasreview Reading in the order the logic runs is the right idea. I would add risk order. A 40 file PR usually has 3 files that really matter: the schema change, the auth check, the retry logic. Show those first with the tests that cover them, and the rest can be skimmed.
@prcanvasreview Reading a change in the order the logic runs makes more sense than alphabetical for big agent PRs. One question: where do files go that the call path never reaches, like config or a migration? Those are the ones that seem easiest to skip when the view follows execution.
@prcanvasreview Alphabetical order explains a lot about the last 20 years of code review. Execution order is a real upgrade. What I'd want on the canvas is the code that didn't change but now gets called differently, since that's where agent PRs hide their surprises.
@prcanvasreview huge agent diffs are mostly mechanical noise: renames, regenerated files, formatting. collapsing all of that into one boring card so the 3 cards with real logic get your attention would be the killer feature. reviewer attention is the scarce resource here, reading speed isn't
@prcanvasreview Tracing by execution path makes way more sense than alphabetical file diffs. How do you handle cross-cutting changes like DB migrations or config updates that don't sit neatly on a call graph?
@prcanvasreview alphabetical files were never the right interface for understanding a change
@prcanvasreview alphabetical file order is an underrated reason agent prs get rubber stamped. reading in the order the change actually flows is how a person would explain the pr anyway
@prcanvasreview is the order from the actual call graph or does an agent guess it
