Repository navigation
fix(dash): drop stale default paths from the gen_cast and gen_screenshots examples - #984
Conversation
Module-level comments in the dashboard crates described things that are not true of the code or the repository. None affects behavior. - Drop the pointers to a wiki/ directory the repository does not have (14 files), keeping the one sentence in each that still carries information. - demo.rs and the two examples named rocm gen-demo, rocm demo, rocm replay and make demo, none of which exist, and default paths under marketing/ and app/. They now name rocm dash --demo / --replay and the session file format. - bench_rollup, rocm-deps, the collectors Cargo.toml, parallel.rs, lib.rs, model_picker, chat, pane and home described a key, a consumer set, a dependency edge, a call path, a tab default or a phase that no longer matches the code. - Remove "Phase N Wave N" labels from the titles of the dash TUI source module headers, together with the sentences that leaned on them, and the pointer to a working-agreements document that is not in the repository. Comment-only change. Fixes #972 Signed-off-by: Jussi Elo <jussi.elo@amd.com>
The rewritten header line ran to about 120 columns; rewrap it to match the surrounding `//!` lines. Signed-off-by: Jussi Elo <jussi.elo@amd.com>
…hots examples `--input` defaulted to /tmp/mi355x-demo.ndjson, which nothing produces any more, and the output defaults pointed into a ../marketing directory that is not part of the repository. Running either example without arguments therefore failed with a confusing "no such file" error or wrote outside the checkout. `--input` is now required, so a missing recording is reported by clap with the flag name. The outputs default to `rocm-dash.cast` and `screenshots` in the working directory, matching the usage shown in each example's header. These are developer examples, not part of the `rocm` CLI, so there is no user-observable CLI behaviour to cover with a scenario. Signed-off-by: Jussi Elo <jussi.elo@amd.com>
There was a problem hiding this comment.
🟢 Approval recommended
No active review issues are reported for the recording-example updates and documentation cleanup.
No open issues 🎉
What changed in this PR
Updates the dashboard’s recording examples to require an explicit input file and write output into the working directory by default. The diff also includes the comment-only dashboard documentation cleanup from the stacked dependency, #978.
Changes:
- Makes
--inputrequired and defaults output torocm-dash.castorscreenshots. - Adds argument-parsing regression tests for both examples.
- Corrects module comments, removes stale documentation pointers, and drops obsolete phase labels without changing dashboard runtime behavior.
| File | Description |
|---|---|
crates/rocm-deps/src/lib.rs |
Corrects the documented dependency consumers. |
crates/rocm-dash-tui/src/ui/update_manager.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/theme.rs |
Removes a stale wiki pointer. |
crates/rocm-dash-tui/src/ui/tabs/pane.rs |
Clarifies how actions open manager overlays. |
crates/rocm-dash-tui/src/ui/tabs/home.rs |
Documents Home as the default tab. |
crates/rocm-dash-tui/src/ui/tabs/chat.rs |
Describes the current consent-gated chat interface. |
crates/rocm-dash-tui/src/ui/services_manager.rs |
Updates the shared execution-pattern description. |
crates/rocm-dash-tui/src/ui/serve_wizard.rs |
Removes obsolete phase terminology. |
crates/rocm-dash-tui/src/ui/runtime_manager.rs |
Removes obsolete phase terminology. |
crates/rocm-dash-tui/src/ui/onboarding.rs |
Updates the first-run flow description. |
crates/rocm-dash-tui/src/ui/model_picker.rs |
Clarifies that preferred engines are displayed, not applied. |
crates/rocm-dash-tui/src/ui/logs_view.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/job_console.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/install_manager.rs |
Removes obsolete phase terminology. |
crates/rocm-dash-tui/src/ui/folder_browser.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/exec.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/examine_manager.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/engine_manager.rs |
Updates shared job-console documentation. |
crates/rocm-dash-tui/src/ui/dock.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/config_manager.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/command_screen.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/automations_manager.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/src/ui/approval.rs |
Removes an obsolete label and unavailable document reference. |
crates/rocm-dash-tui/src/tool_exec.rs |
Describes the current execution methods. |
crates/rocm-dash-tui/src/llm.rs |
Removes an obsolete phase reference. |
crates/rocm-dash-tui/src/jobs.rs |
Removes an obsolete phase label. |
crates/rocm-dash-tui/examples/gen_screenshots.rs |
Requires input, defaults to screenshots, and adds parsing coverage. |
crates/rocm-dash-tui/examples/gen_cast.rs |
Requires input, defaults to rocm-dash.cast, and adds parsing coverage. |
crates/rocm-dash-daemon/src/demo.rs |
Corrects documented demo and replay commands. |
crates/rocm-dash-core/src/vram.rs |
Removes a stale wiki pointer. |
crates/rocm-dash-core/src/state.rs |
Removes a stale wiki pointer. |
crates/rocm-dash-core/src/protocol.rs |
Removes a stale wiki pointer. |
crates/rocm-dash-core/src/partition.rs |
Removes a stale wiki pointer. |
crates/rocm-dash-core/src/metrics.rs |
Removes stale wiki pointers. |
crates/rocm-dash-core/src/lib.rs |
Replaces a stale pointer with a reducer description. |
crates/rocm-dash-core/src/efficiency.rs |
Removes a stale wiki pointer. |
crates/rocm-dash-core/src/bench_schema.rs |
Removes stale wiki pointers. |
crates/rocm-dash-core/src/bench_rollup.rs |
Corrects the documented grouping key. |
crates/rocm-dash-collectors/src/vllm_prom.rs |
Removes a stale wiki pointer. |
crates/rocm-dash-collectors/src/vllm_log.rs |
Removes stale wiki pointers. |
crates/rocm-dash-collectors/src/parallel.rs |
Corrects the documented scrape-helper usage. |
crates/rocm-dash-collectors/src/docker.rs |
Removes a stale wiki pointer. |
crates/rocm-dash-collectors/src/amd_smi.rs |
Removes a stale wiki pointer. |
crates/rocm-dash-collectors/Cargo.toml |
Corrects a dependency comment; dependencies are unchanged. |
Review effort: Balanced
juhovainio
left a comment
There was a problem hiding this comment.
The change itself (dropping the stale default paths in gen_cast / gen_screenshots, with tests pinning the required --input and the new defaults) looks right, and the comment claims I spot-checked against the code hold up. My blocker is how the PR is packaged.
Blocking: the PR doesn't match its description. The body says this is stacked on #978, stays a draft, and that only the last commit belongs here. In reality it is not a draft, #978 is still open, and the base is main, so the diff is 44 files: roughly 40 of them are #978's comment-only changes (35339361, 66c7a196), which the description never mentions. The actual change here is just gen_cast.rs and gen_screenshots.rs. It is also currently conflicting with main. Per the repo's stacked-PR rules, please either mark it draft until #978 merges, retarget the base at #978's branch, or rebase onto main once #978 lands so this PR carries only its own commit.
Non-blocking
- Two small inline comments below (stale "for marketing" wording; a redundant dev-dependency that belongs to #978).
- Output defaults now write into the working directory and are not git-ignored (you flagged this); worth a
.gitignoreentry or a note in the doc comment. - Making
--inputrequired breaks anyone still relying on the old default path. I couldn't confirm there are no other users in docs or scripts, so a quick grep before merge would be good.
CI: only readthedocs reports on this head (passing); no Actions checks are listed, likely because of the merge conflict, so I can't say it's green.
| @@ -35,14 +39,13 @@ use rocm_dash_tui::ui; | |||
| about = "Render TUI views to SVG for marketing" | |||
There was a problem hiding this comment.
about still says "for marketing" while the doc comment and defaults no longer target marketing/. Worth rewording to match, e.g. "Render TUI views to SVG".
There was a problem hiding this comment.
Thanks, fixed in 769f0c3: the about line of gen_screenshots now reads "Render TUI views to SVG". gen_cast already described what it does without naming a purpose, and neither example mentions marketing any more.
| # followed by a chmod hits. Dev-dependencies are exempt from the layering | ||
| # allowlist, so the collectors keep no normal dependency on `rocm-core`. | ||
| # followed by a chmod hits. `rocm-core` is also a normal dependency above | ||
| # (`amd_smi.rs` uses `device_readable_bounded`), so this entry only restates it. |
There was a problem hiding this comment.
The comment now says this dev-dependency "only restates" the normal dependency. If it really is a no-op, the rocm-core dev-dependency line below could be dropped instead of documented. (This belongs to #978's changes, so it's another sign this PR should carry only its own commit.)
There was a problem hiding this comment.
That line belongs to #978, which has merged, so it is no longer part of this PR: after merging main the diff is only the two examples. I checked main: the rocm-core = { path = "../rocm-core" } entry in the dev-dependencies is identical to the normal dependency a few lines above, so it is a no-op, as the comment already says. I left it alone to keep this PR to the two examples; dropping that one line would be a separate cleanup.
There was a problem hiding this comment.
This is now fixed in c21ebdc, replacing my earlier answer that it was out of scope. The dev-dependency entry pointed at the same path as the normal dependency and carried no features, so it was a no-op: I removed the line together with the comment that explained it. I checked that the resolved dependency graph and the features of rocm-core are the same before and after, that Cargo.lock is unchanged, and that the two tests that plant a fake amd-smi still pass through the normal dependency.
nowycondro
left a comment
There was a problem hiding this comment.
🤖 Automated review · review-code · head e76fe4ac9d4b
Outcome: needs work. The gen_cast/gen_screenshots code change itself is correct and well-tested; the blocker is that this PR's packaging doesn't match its own description.
Blocking
- PR state doesn't match its description. The body says this branch stays a draft, is stacked on #978, and that only the last commit belongs here. Live state says otherwise:
draftisfalse, the base ismain(not #978's branch), the diff is 44 files because #978 (commits35339361,66c7a196) is still open and unmerged, andmergeable_stateisdirty(conflicting withmain). One consequence: no Actions CI (build/test/clippy) has run against this head — only thereadthedocscheck reports, and it's unrelated to the Rust change. Fix: mark the PR draft until #978 merges, or retarget its base onto #978's branch, or rebase ontomainonce #978 lands so this PR carries only its own commit. (Confidence 95 — verified via live PR/commit queries, not from the description.)
Verified correct, no issues found
- In
crates/rocm-dash-tui/examples/gen_cast.rsandgen_screenshots.rs: removingdefault_valuefrominput: PathBufis correctly enforced as required by clap's derive; the new unit tests correctly assert theMissingRequiredArgumenterror when--inputis omitted and the new default (rocm-dash.cast/screenshots) once it's supplied. Pattern is applied consistently across both files, and a repo-wide search found no other reference to the old/tmp/...or../marketing/...defaults that this change would break.
Decisions for the author
- Untracked output files (tradeoff, already raised in the PR body): the new defaults write
rocm-dash.cast/screenshots/into the CWD and neither is.gitignored. Worth a maintainer decision either way before merge.
What this covered: the PR's own commit diffed against its stacked parent, both example files read in full (plus an independent second blind pass), repo-wide impact search for the old defaults, and live CI/mergeable-state checks. cargo test/clippy were not re-run here (no Rust toolchain in this environment) — correctness was verified by reading, not execution. Not reconciled against the existing human review thread on this PR.
Not a human review. This is an automated code review; please verify findings independently before acting on them.
Signed-off-by: Jussi Elo <jussi.elo@amd.com>
…se it no longer has The about line still said it rendered views for marketing, while its output default no longer points at a marketing folder. Signed-off-by: Jussi Elo <jussi.elo@amd.com>
|
Caution 3 potential secret finding(s) were committed in this pull request. Treat each credential as compromised and revoke or rotate it now, even if a later commit removed it. Deleting the line or force-pushing does not undo the leak: the commits stay reachable from this pull request and from every clone or fork, and automated scanners harvest secrets pushed to public repositories within minutes. If a finding is a test fixture or a false positive, allowlist it in the configuration of the scanner that reported it. Security scan found 3 critical or high finding(s) on lines changed by this pull request
View the security scan run for complete reports. “New” means the finding is located in a changed file on an added or modified line; it is not a base-versus-head findings delta. |
…heir help gen_cast and gen_screenshots now write rocm-dash.cast and screenshots/ into the working directory by default, which showed up as untracked files after a run from the repository root. Ignore both at the repository root, anchored so no other path of those names is masked, and say in each argument's doc comment that the default is git-ignored at the repository root and that a run from another directory leaves the output there. Signed-off-by: Jussi Elo <jussi.elo@amd.com>
|
The untracked-output finding raised in both reviews is addressed in a1790ed: |
The entry only restated the normal dependency on the same path with no features, so the tests that plant a fake amd-smi reach rocm-core::fake_executable through the normal one. Dropping it leaves the resolved dependency graph, its features and Cargo.lock unchanged, and removes the comment that explained why it was there. Signed-off-by: Jussi Elo <jussi.elo@amd.com>
|
For the review as a whole: the diff against main is now only the two example files, the |
nowycondro
left a comment
There was a problem hiding this comment.
🤖 Automated review · review-code · head c21ebdc6
Review — no blocking findings
Full review of the whole change (4 files: .gitignore, crates/rocm-dash-collectors/Cargo.toml, crates/rocm-dash-tui/examples/gen_cast.rs, crates/rocm-dash-tui/examples/gen_screenshots.rs), diffed against its true merge-base with main rather than the PR description's prose claim.
No ticket identifier (ROCMAI-xxx/EAI-xxx or GitHub issue) is named in the PR description; this is a direct follow-up to review feedback on #978 (merged), which the description accounts for.
Blocking
None
Non-blocking
- Doc-comment wording is slightly ambiguous —
crates/rocm-dash-tui/examples/gen_cast.rs:56-57,crates/rocm-dash-tui/examples/gen_screenshots.rs:45-46
"...is git-ignored at the repository root, so a run from another directory leaves the output there." can be misread as "the ignore rule still applies there" rather than the intended "a run elsewhere writes the file outside the root-anchored ignore rule's reach." Confidence 70 · mechanical · Fix: reword, e.g. "...git-ignored only when produced at the repository root; a run from elsewhere writes the output there instead, uncovered by that rule."
Decisions for the author
- Branch history was merged from
main, not rebased, after the stacked dependency (#978) landed — tradeoff
The repo's ownAGENTS.md(§11, "Stacked and dependent PRs") says to rebase once a dependency merges. The author deliberately merged instead ("to avoid a force-push over a reviewed branch"), and the PR description states it will land as a single squash commit, somain's eventual history is unaffected either way. Confirm this is an accepted exception to the stated convention rather than an oversight, and that squash (not merge-commit) is in fact how this lands. - GitHub's
reviewDecisionis stillCHANGES_REQUESTEDandmergeStateStatusisBLOCKED, even though the packaging problem that review flagged (draft/stacked-base claim not matching live state, 44-file diff from an unmerged #978) is resolved as of this head — verified independently:draftisfalse, base ismain, diff against the true merge-base (7da776cb4) is exactly the 4 files above (35 additions / 15 deletions), andmergeableisMERGEABLE. This is a process/branch-protection state, not a code defect — it just means the stale review needs to be re-requested or dismissed before merge can proceed.
Positive signals
- The new unit tests in both examples assert the actual clap behavior change end to end (
MissingRequiredArgumentwhen--inputis omitted, and the exact new default once it's supplied), not just the struct shape. .gitignoreanchoring (/rocm-dash.cast,/screenshots/) is correctly root-scoped; confirmed no tracked path at the repo root collides with either name.- Repo-wide search at this head found no remaining reference to the removed
/tmp/mi355x-demo.ndjsonor../marketing/...defaults, and no doc/script/CI workflow depends on the old behavior.
Deployment notes
--inputis now a required flag on both dev-only examples (gen_cast,gen_screenshots); any local script or muscle-memory invocation that relied on the old default path will now fail fast with a clear "missing required argument" instead of silently reading nothing or erroring on a stale path. Confirmed these examples are not part of the shippedrocmCLI and are not referenced elsewhere in the repo.- Three Security-scan (PR) checks on this head show
FAILURE(gitleaks, trivy, and the aggregate "Security scanners" job). Inspected the logs directly: gitleaks is flagging 3 "generic-api-key" matches inapps/rocm/src/runtime_install_finalize.rsandapps/rocm/src/serve_summary.rsat commit17f655d3f9...(an already-merged, unrelated commit now reachable because this branch mergedmainin — not one of this PR's own commits, and not a file this PR touches); trivy is flagging pre-existing Dockerfile hygiene items (root user, exposed port 22, no healthcheck). Neither is caused or worsened by this change, so they are not reported as blocking findings here, but they remain genuinely red on this head and will need separate attention (very plausibly pre-existing onmainitself).
What this covered
Every file in the 4-file diff read in full against the PR's actual merge-base (7da776cb404f7cf20037c39676a3ef27f58092ce) and head (c21ebdc6e6fc6b4acb2ef3d507165efcd66e8e4d), via a local read-only clone (PR head fetched to a non-checked-out ref, never checked out). Coupled-file code pass (the two examples + .gitignore) ran as an independent fan-out worker; the Cargo.toml dev-dependency-removal pass was run by the driver directly after one fan-out worker slot failed on a concurrency limit (recorded here rather than silently absorbed: that pass was not run as an independent worker, though it was performed blind, before reading the examples worker's conclusions). Also checked: git history of all four files, the root AGENTS.md for applicable conventions (DCO/signing, stacked-PR rules, the Gherkin-scenario requirement and its dev-example exemption pattern, architecture-doc and dependency-edge checks), all prior review threads and their resolutions, and live CI via gh pr view --json statusCheckRollup. cargo metadata/cargo tree/cargo test were not re-run (no Rust toolchain in this environment); the dependency-graph-equivalence claim was verified by reading Cargo.toml and the workspace root manifest instead of executing.
Not reconciled against the existing human and bot review threads on this PR beyond identifying which of their raised points are now resolved at this head.
Not a human review. This is an automated code review; please verify findings independently before acting on them.
…gnored The help text said the default output is git-ignored at the repository root, so a run from another directory leaves it there, which can be read as the ignore rule still applying elsewhere. It now says the output is git-ignored only when produced at the repository root, and that a run from elsewhere writes it there instead, uncovered by that rule. Signed-off-by: Jussi Elo <jussi.elo@amd.com>
|
The wording note of the latest automated review is addressed in f3321e3: the help text of |
nowycondro
left a comment
There was a problem hiding this comment.
🤖 Automated review · review-code · head f3321e3b
Review — no blocking findings
Review of changes since the review at c21ebdc6, plus the files they touch. The only delta is one new commit (f3321e3b), a 4-line doc-comment wording change in crates/rocm-dash-tui/examples/gen_cast.rs and gen_screenshots.rs; .gitignore and crates/rocm-dash-collectors/Cargo.toml are byte-identical to the previously-reviewed commit. No ticket identifier (ROCMAI-xxx/EAI-xxx or GitHub issue) is named in the PR description; this remains a direct follow-up to review feedback on #978 (merged), which the description accounts for.
Blocking
None
Non-blocking
None
Decisions for the author
- Branch history was merged from
main, not rebased — tradeoff, unchanged since the last pass. The author's comment confirms this is deliberate (avoiding a force-push over reviewed history) and that the project's merge queue squashes each PR to a single commit onmainregardless (citing #978, #982, #985 as precedent). Still worth the author's final confirmation that this is the intended landing path. mergeable_stateisblocked, independently re-verified at this head: no approving review exists (juhovainio's only review is stillCHANGES_REQUESTED, filed against the much earlier commite76fe4ac, and has not been updated or dismissed even though its packaging concern was resolved two commits ago); separately, theSecurity scan / Security scannerscheck is stillfailureon this head (see Deployment notes). Either is sufficient to explainblocked; this is a process/branch-protection state, not a defect in this PR's own commits.
Positive signals
- The doc-comment reword in both examples correctly resolves the ambiguity flagged in the prior automated review: it now states plainly that the root-anchored
.gitignorerule covers the default only when produced at the repository root, and that a run from elsewhere writes the file there, outside that rule's reach — matching the actual/rocm-dash.castand/screenshots/entries added to.gitignore. Confirmed by reading both files at this head and the.gitignorediff against the merge-base.
Deployment notes
- Re-verified live:
Security scan / Security scanners(gitleaks + trivy) is stillfailureon this head. Gitleaks is still flagging the same 3generic-api-keymatches inapps/rocm/src/runtime_install_finalize.rsandapps/rocm/src/serve_summary.rs, all at the already-merged commit17f655d3f9...reachable via this branch's merge frommain— not one of this PR's own commits, and not a file this PR touches. All 46 other checks (build/test/clippy/E2E across platforms, CodeQL, readthedocs) are green on this head.
What this covered
The single-commit delta since the prior automated review at c21ebdc6 (git diff c21ebdc6e6fc6b4acb2ef3d507165efcd66e8e4d f3321e3bfb0c785cce698d83638fb295c0bb587a, confirming only the two example files' doc comments changed), read in full at the new head, plus independent re-verification of live PR state: pull_request_read get (description, mergeable_state), get_reviews (confirming no approving review and the stale CHANGES_REQUESTED), get_check_runs/get_status (CI), and get_comments (confirming the author's own account of what changed and why). The wording change is comment-only, so no test re-execution was needed or performed (no Rust toolchain in this environment regardless). Narrowing basis: the prior review was my own, covered exactly these two files, and git diff against that exact commit shows the only change is this wording — satisfying the attribution/coverage/currency test for narrowing.
Not reconciled against the discussion threads beyond identifying which previously-raised points this commit and the author's comments address.
Not a human review. This is an automated code review; please verify findings independently before acting on them.
juhovainio
left a comment
There was a problem hiding this comment.
Looks good. I found nothing blocking.
Two small notes, neither needs to change:
- Dropping the
rocm-dash-collectorsdev-dependency is unrelated to the example default paths. The PR body says so and Cargo.lock is unchanged, so it's harmless, but it is a separate logical change. - The new output defaults are relative to the working directory and git-ignored only at the repo root. The help text says this, so I think the tradeoff is fine.
CI: the Security scan jobs (gitleaks, trivy) are red. The gitleaks hits are in runtime_install_finalize.rs and serve_summary.rs, which this PR doesn't touch, so they probably come from commits merged from main. Worth confirming that main shows the same failures and that this PR isn't the cause before merging.
This is based on reading the diff; I didn't run the tests locally.
|
On the question about the red security scan: it is not caused by this pull request. The gitleaks findings are the same three |
r0x0r
left a comment
There was a problem hiding this comment.
Approving this. It's small and self-contained, and rather than taking the claims on trust I went and checked them.
Every default this removes really is dead: a repo-wide search turns up nothing that depends on /tmp/mi355x-demo.ndjson or either of the ../marketing/* paths.
Both new tests genuinely fail if you revert the change. I replicated the old and new Args under clap 4.6.7 and confirmed that Old::try_parse_from(["gen_cast"]) returns Ok, so the .err().expect(...) panics on the old code. Each test also pins two separate things — that --input is required, and the exact new default — so neither half of the change can be mutated away while the test still passes.
The ETXTBSY rationale in the deleted Cargo.toml comment isn't lost: it's spelled out in full in the module docs of crates/rocm-core/src/fake_executable.rs, and rocm-core is already a normal dependency with an identical spec, so removing the entry really is a no-op.
I also ran the two things that are easy to get wrong by inspection: cargo test --all-targets does pick up tests in example targets, and CI's heavy filter matches **/*.rs, so the Windows lane runs these.
On the three red security checks — none of them are this PR's doing. Trivy's HIGH is DS-0002 on tests/remote-ssh/Dockerfile, which this PR doesn't touch. The gitleaks hits are commit messages from #910 that are already on main; they only surfaced here because merging main widened scan_mode: changed to cover main's content. Every required check is green, and a squash or rebase would quiet the rest.
One question inline about the working-directory default, but it's a question, not a blocker.
| /// to the working directory. It is git-ignored only when produced at the | ||
| /// repository root; a run from elsewhere writes the file there instead, | ||
| /// uncovered by that rule. | ||
| #[arg(long, default_value = "rocm-dash.cast")] |
There was a problem hiding this comment.
Just confirming the reasoning here rather than asking for a change — the PR body says this was a deliberate call, so I expect the answer is "yes, on purpose."
The new default writes into the working directory, which does match the usage each example documents. The cost is two .gitignore entries at the repo root for what are really developer-tool artifacts. The alternative would be making --output (and --output-dir in gen_screenshots.rs) required, symmetric with what you've just done to --input, which would need no .gitignore change at all.
Was the convenience of a default deliberately preferred over that symmetry?
Summary
Follow-up to the review on #978 (merged), which noted that the corrected
--helptext of thegen_castandgen_screenshotsdashboard examples is accurate but their defaults still point at things that do not exist.--inputdefaulted to a fixed/tmppath that nothing produces any more. It is now required, so a missing recording is reported by clap with the flag name instead of a confusing "no such file".../marketing/...directory that is not part of this repository, so a default run wrote outside the checkout. They are nowrocm-dash.castandscreenshotsin the working directory, matching the usage shown in each example's header.--inputis required (the clapMissingRequiredArgumenterror) and that the output default is the new exact value.aboutline ofgen_screenshotsno longer says "for marketing", since its output no longer points at a marketing folder.rocm-coreentry under[dev-dependencies]ofrocm-dash-collectorsis dropped together with its comment. It pointed at the same path as the normal dependency and carried no features, so it was a no-op that the earlier stacked change left behind (a review of that change asked for it to be removed rather than documented).This branch was stacked on #978. #978 has merged, so
mainis merged into the branch and the diff againstmainis now only the two example files (gen_cast.rsandgen_screenshots.rs), the two.gitignoreentries and the small manifest cleanup. The branch history still lists the pre-merge commits of the earlier change, becausemainwas merged in rather than rebased to avoid a force-push over a reviewed branch; the pull request lands as a single squash commit.Decision on the output location
The new defaults write into the current working directory. Both reviews that raised this asked for it to be settled before merge, and it is now:
/rocm-dash.castand/screenshots/are added to the root.gitignore, anchored at the repository root so no other path of those names is masked, and the doc comment of each output argument says that the default is git-ignored at the repository root and that a run from another directory leaves the output there. Nothing tracked is affected: no tracked file matches either entry, and a same-named path below another directory still shows as untracked.Verification
cargo test -p rocm-dash-tui(1086 library tests and the other targets),cargo test -p rocm-dash-tui --examples(the two new tests pass), a check that the two outputs created in the repository root stay out ofgit status,cargo clippy -p rocm-dash-tui --all-targets -- -D warnings,cargo fmt --checkandcargo xtask check-architecture-docall pass.rocm-coreforrocm-dash-collectorsare the same before and after (cargo metadataandcargo tree, which differ only by one duplicate listing of the same node),Cargo.lockis unchanged, andcargo build,cargo test(140 library tests, including the two that plant a fakeamd-smi),cargo clippy -D warnings,cargo fmt --checkandcargo xtask check-crate-edgespass forrocm-dash-collectors.--inputdefault in either example turns its test red (checked again after the merge, and restored byte-identical).scripts/smoke_local.py, and the two examples end to end.These are developer examples, not part of the
rocmCLI, so there is no user-observable CLI behaviour and no Gherkin scenario applies. A search of the whole repository found no document, script or workflow that refers to the examples or to their old default paths; the only mention of an example file is a fixture list in thextaskarchitecture-doc test.