Profile dashboard: the datasets tab as a filterable list, and a draft/published lifecycle for Datasets

Aperta
#2,614 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 3 giorni

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
22/100
Tipo di issue
Funzionalità
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
django, python
Ambito
full-stack

Direzione di ricerca

This is an epic, not a first issue: start by reading the existing tables tab (the list the datasets tab must mirror) and the Dataset DRF views/serializers behind the public API. Map the user stories into workstreams — draft/published lifecycle, filterable list, members drawer, delete name-holding, API parity — and pick one vertical slice. Done means each slice has its own issue with named files, a test, and a maintainer sign-off; nothing here is scoped to a single PR.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

datasets Profile ready-for-agent

Problem Statement

A user who curates data on the Open Energy Platform groups Tables into Datasets: a named
collection of Tables, theirs or anyone's published ones, described by one oemetadata document. Today
the datasets tab of their profile dashboard, and the Dataset itself, let them down in three ways.

There is no draft. Every Dataset is public the moment it is created. It appears in the public
topic list, on its detail page, in the table page's Datasets sidebar and in the public API, which
returns every Dataset on the platform to anonymous callers, unpaginated, with every member's full
schema. A user cannot prepare a Dataset privately and publish it when it is ready. The lifecycle was
promised with the first Dataset UI (#1971) and never built.

The tab is not a working list.

  • Datasets are large cards, eight per page, ordered by creation date only.
  • The only filter is a substring search. A user cannot ask "which of my Datasets are still drafts?",
    "which ones are about grid?", or "which ones hold Tables tagged wind?".
  • There is no modification date at all.
  • Nothing can be done to several Datasets at once.

Managing members is cramped and inconsistent.

  • The inline manage panel's Table picker stops at 20 results, while the tables tab can add up to
    2,500 Tables to a Dataset in one go.
  • The tables tab's "Add to dataset…" refuses Tables the user holds no role on, while the Dataset's
    own picker accepts any published Table. The two sides follow two different rules.
  • Delete has no confirmation of consequences, no log, and frees the Dataset's name at once, even
    while scenario bundles in the knowledge graph still cite it by that name. Whoever takes the name
    next silently becomes the target of those citations.

The API cannot do what the dashboard does. It can neither read nor set a Dataset's Topics, so a
Dataset whose members carry no Topic could never be published through it. Its PUT erases the
stored @id when omitted. Its DELETE answers 204 with a body. Refusals use three different
body shapes.

Solution

A Dataset gets a lifecycle. It starts as a draft Dataset, visible only to its creator, and
becomes a published Dataset when the creator publishes it. Publishing is allowed when the
Dataset passes its Publish gate: at least one member Table and at least one Topic. Unpublishing
is always allowed. Every place on the platform that shows a Dataset to someone follows one rule:

  • the public catalogue lists published Datasets only;
  • lists beside your own work show published Datasets plus your own drafts;
  • a single draft answers 404 to everyone but its creator;
  • a scenario bundle citing a draft cannot tell whether the link is alive.

Existing Datasets become published Datasets, so nothing public goes dark on deploy.

The datasets tab becomes the tables tab's sibling. One dense, sortable, filterable list of the
user's own Datasets, 25 per page, with the address bar holding the state:

  • Columns: ☐ · Dataset · Status · Tables · Topics · Modified · Created · ⋯.
  • Filters: search, a status segment with counts, Topic, Tag, and Created and Modified ranges.
  • Row ⋯ menu: Edit…, Manage tables…, Publish… or Unpublish…, Delete….
  • Bulk bar: publishes, unpublishes or deletes many Datasets at once.

Create and Edit are one dialog that shows the Dataset's future web address while the user types the
title.

Members are managed in a drawer.

  • It shows every member, paged and searchable, and adds any Table the user may assign, with one
    click per Table.
  • A link hands over to the tables tab, filtered to this Dataset, for bulk work on the user's own
    members. Browser Back returns to the drawer.
  • One rule decides what may be added, on both tabs and in the API.

A deleted Dataset's name is free again unless a scenario bundle cites it. While cited, it is held
for 30 days and then released by an operator command that removes the dead citations.

The API can do everything the dashboard does, through the same service.

  • New: read and set Topics, PATCH instead of PUT, publish and unpublish routes, ?mine=true and
    ?published= filters.
  • Paginated summaries replace the unpaginated full list.
  • Refusals use DRF's body shape throughout, and the OpenAPI description is checked against real
    responses.

User Stories

Seeing my Datasets
  1. As a Dataset creator, I want all my Datasets in one dense list instead of cards, so that I can scan them at a glance.
  2. As a Dataset creator, I want 25 Datasets per page with a pager, so that a long list stays usable.
  3. As a Dataset creator, I want each row to show the title with the fixed technical name beneath it, so that I recognise the Dataset and its web address.
  4. As a Dataset creator, I want the title to open the Dataset's detail page in the same tab, so that Back brings me to the same list state.
  5. As a Dataset creator, I want a Status badge reading Draft, or Published with "since ‹date›", so that I see at once what is public and since when.
  6. As a Dataset creator, I want a Tables cell that counts the members, so that I see how big each Dataset is.
  7. As a Dataset creator, I want the Tables count to open a list of up to ten member titles, each linked and marked Draft or Embargoed, so that I can check the contents without leaving the list.
  8. As a Dataset creator, I want that popover to summarise how many members are drafts or embargoed, in neutral words, so that I know what others will not see without it reading as an error.
  9. As a Dataset creator, I want "and n more" in the popover to open the members drawer, so that I can reach every member of a large Dataset.
  10. As a Dataset creator, I want a draft's empty Tables or Topics cell to say "– needed to publish", so that I see what stands between the draft and publishing.
  11. As a Dataset creator, I want a published Dataset's empty Tables or Topics cell to show a plain "–", so that a published Dataset that lost its members is not presented as an error.
  12. As a Dataset creator, I want up to two Topic chips and a "+n" that opens the rest, so that the row stays one height.
  13. As a Dataset creator, I want a Modified date that changes when I edit the title, description or Topics, or when the membership changes, so that I can find what I worked on recently.
  14. As a Dataset creator, I want Modified to stay put when someone changes the rows inside a member Table, so that my sort is not driven by strangers' activity.
  15. As a Dataset creator, I want Modified to stay put when I publish or unpublish, so that the date means "content changed".
  16. As a Dataset creator, I want a Dataset older than the Modified field to show "–" with an explanation for screen readers, so that no date is invented.
  17. As a Dataset creator, I want a Created date on every row, so that I can tell old Datasets from new ones.
  18. As a Dataset creator, I want the list to show only Datasets I created, so that the tab is about my own work.
  19. As a Dataset creator with no Datasets, I want a short explanation of what a Dataset is and a "Create a dataset" button instead of an empty table, so that I know how to start.
Filtering and sorting
  1. As a Dataset creator, I want to search title, technical name and description, so that I find a Dataset by any of them.
  2. As a Dataset creator, I want a status segment All · Draft · Published with counts that follow my other filters, so that I see how many of each match.
  3. As a Dataset creator, I want a segment showing 0 to be greyed but still clickable, so that the counts never trap me.
  4. As a Dataset creator, I want to filter by Topic, matching any of the Topics I tick, so that I find Datasets in an area.
  5. As a Dataset creator, I want the Topic filter to offer only Topics that occur on my Datasets, so that every option can match.
  6. As a Dataset creator, I want to filter by Tag, where a Dataset matches when one of its member Tables carries the tag and several tags must all be found, so that I find collections about a subject the same way the public dataset list does.
  7. As a Dataset creator, I want the Tag filter to offer only tags carried by members of my Datasets, so that every option can match.
  8. As a Dataset creator, I want Created and Modified date ranges behind "More filters", either end optional, so that I can narrow by time without crowding the bar.
  9. As a Dataset creator, I want a Dataset whose Modified is unknown to match no Modified range, so that a range never pretends to know.
  10. As a Dataset creator, I want every active filter shown as a removable chip with "Reset all", so that I can see and undo what narrows the list.
  11. As a Dataset creator, I want Reset to put Status back to All and keep my sort, so that resetting does not lose the order I chose.
  12. As a Dataset creator, I want "No datasets match these filters" with a Reset button when nothing matches, so that I can get back quickly.
  13. As a Dataset creator, I want to sort by Dataset title, Status, Tables count, Modified and Created, each both ways, so that I can order the list as I need.
  14. As a Dataset creator, I want the Status sort to order published Datasets by how long they have been published, drafts at one end, so that "recently published" is one click.
  15. As a Dataset creator, I want the default order to be newest Modification first, so that a Dataset I just created or changed comes to the top.
  16. As a Dataset creator, I want Datasets with no known Modified date at the end in both directions, so that unknown dates do not crowd the top.
  17. As a Dataset creator, I want ties broken by title and then a fixed key, so that paging never shows a row twice or skips one.
  18. As a Dataset creator, I want the filters, sort and page in the address bar, so that I can bookmark, share and come back to a view.
  19. As a Dataset creator, I want a changed filter to take me to page 1, so that I do not land on an empty page.
  20. As a Dataset creator, I want an address with a value that no longer applies (a Topic no Dataset has any more) to show it as a muted "no longer applies" chip instead of failing, so that old links still work.
  21. As a Dataset creator, I want filters to apply as I type or choose, without a submit button, so that the list follows my intent.
Creating and editing
  1. As a Dataset creator, I want a "New dataset" button in the list header, so that creating is always one click away.
  2. As a Dataset creator, I want Create and Edit in one dialog with title, description and Topics, so that I learn one form.
  3. As a Dataset creator, I want to see the web address derived from my title while I type, so that I know the permanent name before I commit.
  4. As a Dataset creator, I want to be told before saving that a name is already taken, so that I change the title instead of failing on save.
  5. As a Dataset creator, I want to be told before saving that a name is held because scenario bundles still cite a deleted Dataset of that name, so that I understand why it is not available.
  6. As a Dataset creator, I want to be told when a title contains no letter or number to make a name from, so that I fix it before saving.
  7. As a Dataset creator, I want the name shown as fixed in the Edit dialog, so that I know editing the title does not move the address.
  8. As a Dataset creator, I want Topics optional at creation, with a note that at least one is needed to publish, so that I can start a draft quickly.
  9. As a Dataset creator, I want an unknown Topic to be refused rather than silently dropped, so that what I save is what I chose.
  10. As a Dataset creator, I want a new Dataset to start as a private draft, so that I can prepare it before anyone sees it.
  11. As a Dataset creator, I want the members drawer to open on my new Dataset right after Create, with the cursor in the add search, so that the obvious next step is ready.
  12. As a Dataset creator, I want a save that changes nothing to write nothing and leave Modified alone, so that opening and closing the form is harmless.
  13. As a Dataset creator, I want a refused save to keep what I typed and say why, so that I do not retype it.
Managing member Tables
  1. As a Dataset creator, I want "Manage tables…" to open a drawer listing every member, 25 per page, with a search over members, so that I can manage a Dataset of thousands of Tables.
  2. As a Dataset creator, I want each member marked Draft or Embargoed where it is, so that I see what others will not see.
  3. As a Dataset creator, I want to remove a member with one click and no dialog, so that curating is quick.
  4. As a Dataset creator, I want to be asked first when removing a member I could not add back (someone else's draft or embargoed Table), so that I do not lose it by accident.
  5. As a Dataset creator, I want to remove any member, whoever holds it, so that I control what my Dataset contains.
  6. As a Dataset creator, I want to be told, without being stopped, that a published Dataset whose last member I removed stays published, so that I can decide to unpublish it.
  7. As a Dataset creator, I want an add search over every Table I may assign, paged rather than capped, so that I can find any of them.
  8. As a Dataset creator, I want the add search to list my own Tables before I type, so that the most common addition needs no typing.
  9. As a Dataset creator, I want to add any published Table that is not under an embargo, whoever holds it, so that I can curate others' data.
  10. As a Dataset creator holding Data editor or more on a draft or embargoed Table, I want to add it to my Dataset, so that I can stage a release before publishing.
  11. As a Dataset creator, I want an Add button per result that turns into "Added" in place, so that I can add several in a row.
  12. As a Dataset creator, I want a result already in the Dataset shown as "Already in", so that I do not try twice.
  13. As a Dataset creator, I want to be told when adding a Table also added its Topics to my Dataset, so that I understand why my Dataset's Topics changed and why it may now be publishable.
  14. As a Dataset creator, I want a Topic I removed to stay removed when I add more Tables later, so that my curation of Topics survives.
  15. As a Dataset creator, I want a link "Work on your own tables in this dataset in the tables tab" with a count of how many members are mine, so that I can publish, share or remove several of my own members at once.
  16. As a Dataset creator, I want browser Back from that hand-off to bring me back to the list with the drawer open again, so that I continue where I left.
  17. As a Dataset creator, I want the drawer to have its own address, so that I can bookmark or share the members view of my Dataset.
  18. As a Dataset creator, I want the list behind the drawer to refresh after every change without the drawer closing or my focus moving, so that counts and dates stay true while I work.
  19. As a Dataset creator, I want the same rule to decide what may be added whether I add from the drawer, from the tables tab or through the API, so that the three never disagree.
Publishing and unpublishing
  1. As a Dataset creator, I want "Publish…" on every draft row, so that I can publish when ready.
  2. As a Dataset creator, I want Publish on a draft that fails the Publish gate to name what is missing ("Add at least one table", "Choose at least one topic"), with links to Manage tables… and Edit…, so that I can fix it in one step.
  3. As a Dataset creator, I want the publish dialog to tell me how many members are drafts or embargoed, as information only, so that I know they stay hidden until their holders publish them.
  4. As a Dataset creator, I want to publish a Dataset that holds draft or embargoed members, so that a stranger's unpublished Table never blocks my release.
  5. As a Dataset creator, I want a published Dataset to be listed publicly under its Topics, so that others find it.
  6. As a Dataset creator, I want "Unpublish…" on every published row, always allowed, so that I can take a Dataset back.
  7. As a Dataset creator, I want the unpublish dialog to say in words what happens (it leaves the catalogue, others get 404, scenario bundles citing it can no longer tell whether the link is alive) and what stays, so that I decide knowingly.
  8. As a Dataset creator, I want a member Table being unpublished, embargoed or deleted by its holder never to change my Dataset's state, so that my release is not undone by someone else.
Deleting
  1. As a Dataset creator, I want "Delete…" last in the menu, in red, so that it is not hit by accident.
  2. As a Dataset creator, I want the delete dialog to say that the member Tables are not deleted and only leave the Dataset, so that I am not afraid of losing data.
  3. As a Dataset creator, I want a draft deleted with a plain confirmation and a published Dataset only after typing its name, so that the more public deletion takes more care.
  4. As a Dataset creator, I want the delete dialog for a published Dataset to say that it leaves the catalogue, its page and API address answer 404, and scenario bundles citing it will read the link as deleted, so that I know the consequences.
  5. As a Dataset creator, I want the name of a deleted Dataset to become free again when nothing cites it, so that names are not used up.
Acting on many Datasets
  1. As a Dataset creator, I want a checkbox per row, a tri-state header box for the page and Shift-click ranges, so that I can select quickly.
  2. As a Dataset creator, I want "Select all N matching datasets" once the page is selected, so that I can act on everything a filter found.
  3. As a Dataset creator, I want the selection to clear, with a note, when I change a filter, so that I never act on rows I can no longer see.
  4. As a Dataset creator, I want a bulk bar with Publish…, Unpublish… and Delete… that appears in a reserved slot, so that the list does not jump under the pointer on the first tick.
  5. As a Dataset creator, I want bulk publish to leave out drafts failing the gate, each with its reason, and Datasets already published, so that I see exactly what will happen.
  6. As a Dataset creator, I want bulk unpublish to leave out drafts, so that nothing is done twice.
  7. As a Dataset creator, I want bulk delete to ask me to type the number of Datasets when any of them is published or there are more than ten, so that a large or public deletion takes care.
  8. As a Dataset creator, I want a bulk action to apply to all of the Datasets it lists or to none, so that I never end up half done.
  9. As a Dataset creator, I want a bulk action refused because something changed meanwhile to say "Nothing was changed" and show the check again, so that I can retry with the current state.
  10. As a Dataset creator, I want a clear limit on how many Datasets one bulk action takes, so that a request never times out halfway.
After an action
  1. As a Dataset creator, I want the list to refresh itself after every action with my filters, sort and page kept, so that counts and rows are always current.
  2. As a Dataset creator, I want a confirmation that names what was done, and says when an affected row no longer shows under the current filter, so that a row that disappeared is explained.
  3. As a Dataset creator, I want focus to go back to the row's ⋯ button, or to the list heading if the row is gone, so that I keep my place with a keyboard.
  4. As a Dataset creator, I want a logged-out session to tell me so and offer to log in again and return to this view, so that an expired session is not mistaken for an error.
  5. As a Dataset creator, I want a server error to say that the change may not have been made, so that I check rather than assume.
Others, and the privacy of drafts
  1. As a visitor, I want the public topic list to show only published Datasets, so that I see finished collections.
  2. As a visitor, I want a draft Dataset's page, metadata JSON and API address to answer 404, the same as a name that never existed, so that drafts are private and nothing reveals what someone is preparing.
  3. As a Dataset creator, I want to open my own draft's detail page, marked as a draft, so that I can check how it will look.
  4. As a Dataset creator, I want my own drafts to stay out of the public topic list even for me, so that the catalogue looks to me as it looks to everyone.
  5. As a user viewing a Table's page, I want its Datasets sidebar to list the published Datasets it belongs to plus my own drafts, so that I see the collections I may see.
  6. As a user on the tables tab, I want the Datasets column, its filter and the dataset dialogs to show published Datasets plus my own drafts, so that the two tabs agree.
  7. As a Table holder, I want deleting my Table to say nothing about other people's draft Datasets it was in, so that their preparations stay private.
  8. As a platform admin, I want no special view of other people's drafts, so that the privacy rule has no exception to audit.
  9. As a user trying to create a Dataset, I want a name used by someone's draft to read as taken, without revealing anything else, so that names stay unique.
API clients
  1. As an API client, I want each Dataset read to carry published_at (null for a draft), creator, topics and modified_at, so that I know its state, owner, Topics and last Modification.
  2. As an API client, I want to set Topics when creating and to replace them when updating, with unknown names refused naming them, so that a Dataset can pass the Publish gate through the API alone.
  3. As an API client, I want PATCH where an omitted key stays untouched, so that fixing a title does not erase the @id.
  4. As an API client, I want a name in an update refused, so that I learn the name is fixed instead of being ignored.
  5. As an API client, I want POST …/publish/ and POST …/unpublish/, returning the Dataset's body, so that I can release from a pipeline and see the new published_at.
  6. As an API client, I want a publish that fails the gate to answer 409 with the failed checks as a list of names, so that a pipeline can react without parsing prose.
  7. As an API client, I want unpublishing a draft to succeed without writing anything, and publishing a published Dataset to republish it, so that retries are safe.
  8. As an API client, I want DELETE to go through the same service and log as the dashboard and answer a bare 204, so that deletions are recorded and HTTP-correct.
  9. As an API client, I want ?mine=true to list my own Datasets including drafts, and ?published=true|false to narrow by state, so that a pipeline can find its drafts.
  10. As an anonymous API client, I want ?mine=true to answer 401, so that I know authentication is needed rather than seeing an empty list.
  11. As an API client, I want the list paginated with page and page_size and its items to be summaries with a resource count, so that a page stays small however large the members are.
  12. As an API client, I want the resources of a Dataset paginated too, so that a Dataset of 2,500 Tables does not arrive in one response.
  13. As an API client, I want the member routes to answer 404 for a Dataset I cannot see and 403 for one I can see but do not own, so that the API reveals no more than a read.
  14. As an API client, I want a limit of 2,500 Tables per member call with a 400 naming it, so that a request never times out halfway.
  15. As an API client, I want every refusal as {"detail": …}, or DRF's field map for validation, so that one error handler serves the whole Dataset API.
  16. As an API client, I want the OpenAPI description to state each operation's auth expectation and the draft rule in words, and to declare every success and refusal body, so that a generated client handles what it will meet.
Scenario bundles (OEKG)
  1. As a scenario bundle author, I want a link to a draft Dataset to read resolvable: null, the same for every reader, so that a draft is neither leaked nor reported as deleted.
  2. As a scenario bundle author, I want a link to a deleted Dataset to read resolvable: false, so that I can see that my citation is dead.
  3. As a scenario bundle author, I want a cited Dataset's name held while my link exists, so that nobody else's Dataset silently becomes the target of my citation.
Keyboard, screen readers, small screens
  1. As a keyboard user, I want native tab order with every control reachable and focus kept by stable ids across refreshes, so that I can work without a mouse.
  2. As a screen-reader user, I want every icon-only or "–" cell to carry visually hidden text, so that I hear what a sighted user sees.
  3. As a screen-reader user, I want the result count announced from a live region that is never replaced, so that every change is announced reliably.
  4. As a keyboard user, I want popovers that contain links to take focus when opened and give it back on Escape, so that their links are reachable.
  5. As a laptop user with the profile sidebar, I want every column visible down to a 1280 px screen, so that I lose nothing on a common laptop.
  6. As a tablet user, I want Created and then Topics to drop before anything scrolls sideways, so that the row stays readable.
  7. As a phone user, I want each row stacked into two lines (title and ⋯; Status, Tables, Modified) with a "Sort by" select and a "Select this page" box, so that the list works without column headers.
Operators
  1. As an operator, I want one structured log line per Dataset changed (create, edit, publish, unpublish, delete, each member added or removed), naming who, what and through which route, so that changes can be traced.
  2. As an operator, I want a command that, by default, lists the citations to Datasets deleted more than 30 days ago, and with --apply removes them and frees the names, writing its record before it touches the graph, so that held names do not stay held forever.
  3. As an operator, I want the lifecycle migration to make every existing Dataset published since its creation date, so that nothing that was public goes dark on deploy.

Implementation Decisions

Already in place (reuse, do not rebuild)
  • The list mechanics of the tables tab: Filter, ChoiceFilter, RangeFilter, Segment,
    Sort (with nulls=), Listing, ListState (URL state, stale values, canonical links). They
    know no model and work with a uuid key.
  • The generic list partials: segment, sort header, sort select, pager, chips. The filter control
    needs its one Table word made a noun.
  • The cell partials created and modified (as is, given the row interface), and chips
    (given a row key instead of a Table pk).
  • The popover behaviour module, the owner rule for profile/<user_id>/ routes (foreign 404,
    anonymous page 302, anonymous htmx 401, no admin exemption), and the structural owner-rule test.
  • The assignment rule assignable_tables(user) and assign_table (adds the Table and its
    Topics additively, never the draft pseudo-topic).
  • The name rule normalize_dataset_name.
Prerequisite: a behaviour-neutral refactor of the tables tab (one or two slices first)

About 1,100–1,300 lines moved or touched, proven by the unchanged tables-tab suites:

  • Page and region templates get a base with blocks for ids, events and copy.
  • The tab's JS becomes a generic bindList(doc, config) module. bindTablesTab stays as a thin
    wrapper, so its existing vitest suite runs unchanged.
  • The CSS splits into generic list styles and per-tab blocks. The column-collapse container
    queries are scoped per tab
    : they hide columns by class name, so a datasets column of the same
    name would otherwise collapse at the tables' widths.
  • The views get base classes with hooks (listing, rows, service, dialog template, event names).
  • The width measurement script gains parameters (URL, column sets, ids, probe query), and the
    benchmark seed gains Datasets.

The shared action vocabulary (left-out groups, refusal exceptions, the preflight core,
noun-parameterised messages) is extracted once the second service exists, not before.

Schema changes
  • Dataset.published_at: a nullable timestamp; null = draft, a value = published since then.
    • The model default is draft, so a bare create is a draft. Fixtures that need a published
      Dataset set it.
    • Republishing overwrites it: it means "published since", not "first published".
    • The migration sets published_at = created_at for every existing Dataset, never the deploy
      time.
    • No separate state field: one field, so state and date cannot disagree.
  • Dataset.modified_at: a nullable timestamp, the last Modification of the Dataset itself.
    • A Modification is a change to its title, description, Topics or membership, from either tab or
      the API, including a member Table being deleted by whoever holds it.
    • Not a Modification: changes inside member Tables, publish, unpublish.
    • New Datasets get modified_at = created_at. Existing Datasets migrate as NULL (shown as
      "–"); nothing is backfilled.
    • Stamped only on a real change, by the service, never in a view.
  • DatasetNameHold: name (primary key) and deleted_at. It exists only while a deleted
    Dataset's name is cited by a Dataset link.
  • No Dataset embargo: an embargo belongs to a member Table. No stored "publishable" flag: the
    Dataset gate runs live (two EXISTS).
The lifecycle rule (one place, on the model, anonymous-safe)
  • Dataset.objects.published(): the public catalogue and OEKG resolution.
  • Dataset.objects.visible_to(user): published plus the user's own drafts; published only for
    an anonymous user. Used by lists beside the viewer's own work.
  • dataset.readable_by(user): single-object reads, and the read step of every API write.
  • No platform-admin exemption anywhere.
  • The tables tab's local PUBLISHED_DATASETS / visible_datasets are removed; callers use the
    queryset.
  • The Dataset Publish gate is declared once, as DATASET_GATE beside the Table's
    PUBLISH_GATE. Its checks are named members and topics. It judges the transition only: a
    published Dataset that later stops passing stays published.
  • A member's state never gates or changes the Dataset's state: a published Dataset may hold, and
    later gain, draft and embargoed members.
  • Ownerless Datasets (creator null) stay frozen: published ones are readable and editable by
    nobody; drafts are visible to nobody.

Every reader:

Reader A draft is visible to Everyone else
Public topic list nobody, the creator included absent
Table page Datasets sidebar, API list, tables tab (column, filter, dialogs) the creator, marked as draft absent
Detail page, its metadata JSON, API detail, API resources the creator, with a draft banner or state 404
OEKG ref: dataset resolution nobody resolvable: null, tables: null (the existing "unknowable")

Writes never tell more than a read.

  • API writes resolve the Dataset through the read rule first: not visible → 404, visible but not
    yours → 403.
  • Dashboard routes answer 404 for any Dataset that is not the user's own.
  • A foreign draft and a name that never existed get the same 404 body.
  • Accepted residual: a draft's name reads as taken on create, because names are global.
  • A Table delete that removes a member from a stranger's draft is silent to the deleter. The
    operator log is the only record.
Dataset action service (a new deep module, shared by the dashboard and the API)

It sits beside the Table action service and follows the same contract:

  • preflight returns what would happen, with left-out groups and their reasons.

  • execute(user, action, names, params, via) re-checks under row locks, refuses the whole
    request or writes everything in one transaction, and logs one line per Dataset after commit on
    oeplatform.dataset_actions:

    dataset_action dataset=… action=… by=… via=dashboard|api batch=…
    

    Delete adds published= and members=; member changes write one line per Table.

Actions:

  • create: through the shared create path. Topics are validated by the service; unknown names
    and the draft pseudo-topic are refused, naming them. A held name is refused like a taken one.
  • edit: title, description, @id, Topics. An edit that changes nothing writes and stamps
    nothing.
  • publish: the gate runs live. A refusal names the failed checks. It sets published_at to
    now. A republish passes the gate again and overwrites the date.
  • unpublish: always allowed. On a draft it is a no-op success.
  • delete:
    • Members are never deleted. The typed confirmation is enforced in execute, under the lock:
      the name for one published Dataset, the count for a batch with any published Dataset or more
      than ten.
    • Before deleting, the service asks the graph store whether any ref: dataset link cites the
      name, in one ASK. Not cited → the name is free at once. Cited, or the store unreachable → a
      DatasetNameHold is written in the same transaction.
      A delete never fails on the graph store.
  • members_add / members_remove: the Dataset is the subject, Tables are the items.
    • Add: eligible iff in assignable_tables(user); already-members are left out ("Already in");
      a Table that is no longer assignable at execute refuses the request whole. It writes through
      assign_table. No Table role gate.
    • Remove: the creator removes any member, whoever holds it. Non-members are left out ("Not
      in"). Topics stay.
    • Ceiling 2,500 Tables per call, the measured tables-tab number. modified_at is stamped only
      when something was actually added or removed.
  • Ceilings for publish, unpublish and delete: one shared number, measured in the first slice
    that adds them. There is no OEDB work, at most one ASK per delete; expected in the hundreds.
    Over the ceiling the preflight reads no Dataset.

Elsewhere:

  • The tables tab's dataset_add / dataset_remove stay on the Table action service (the
    reverse direction: many Tables into one Dataset). They share assign_table and the Modification
    stamp.
  • Every write path that changes membership stamps modified_at: both services, the API member
    routes, and the Table delete path (which removes memberships).
Name holds and the cleanup command
  • While a hold exists, create and the dashboard's name preview refuse the name: "used by a deleted
    dataset that scenario bundles still cite; it becomes available once those links are gone."

    Create reads the hold table, never the graph store.
  • manage.py drop_dead_dataset_links, shaped like repair_oekg_shape_violations:
    • dry run by default, --apply to act, --older-than-days defaulting to 30;
    • for every hold older than that, it removes each ref: dataset link to the name, one atomic
      update with a version bump per bundle, then deletes the hold;
    • its JSON record is written before the graph is touched, and it logs one line per link;
    • it writes no OEKG history entry;
    • an operator runs it on the host; there is no scheduler.
  • This is a deliberate, interim departure from two closed decisions of the OEKG API map: WF-11
    ("never collect") and WF-08 ("every write records"). The full fix is out of scope.
Columns (fixed set, no column chooser)

☐ · Dataset · Status · Tables · Topics · Modified · Created · ⋯

Column Content
☐ selection
Dataset title on line 1, the technical name muted on line 2; the title links to the public detail page in the same tab
Status one badge: Draft, or Published with a second line "since ‹date›"
Tables a count button ("5 tables"); see below
Topics up to two chips and "+n" opening the rest
Modified the date, or "–" with visually hidden text when unknown
Created the date
⋯ the row menu, sticky at the right edge
  • The Tables popover holds one neutral line naming only the non-zero parts ("2 drafts · 1
    embargoed"), then at most ten members by title with Draft or Embargoed badges, then "and n more"
    opening the members drawer.
  • The gate hint. On a draft, an empty Tables or Topics cell reads "– needed to publish" on one
    line, with screen-reader text. On a published Dataset it is a plain "–". If the Dataset gate ever
    gains a check no column shows, a Publishable column is added in the same change.
  • No description on the row. It is searchable, and shown on the detail page and in the edit
    form.
  • No "yours / someone else's" marker on members, in the popover or the drawer.
  • The columns are a per-tab declaration, never a shared template with conditionals, so a later
    Access column (Dataset roles) is one entry plus a re-measure.
  • A shared status-badge partial is extracted from the tables tab's inline badge, with a "since"
    variant. members is a new cell partial.
Filters, URL, paging
Filter Kind Parameter Semantics
Search text search substring of name, title, description
Status segment, faceted counts status draft (published_at null) · published
Topic multi choice topics the Dataset's own Topics, OR
Tag multi choice tags tag keys; a Dataset matches a tag when some member carries it; several tags AND, possibly on different members
Created range, under "More filters" created_from, created_to either end optional
Modified range, under "More filters" modified_from, modified_to either end optional; unknown matches nothing
  • Filters AND together. Tags are read-only: nothing is copied onto the Dataset. Each tag is
    one pk__in subquery over the membership table, never a join.

  • Option sources are scoped to the user's own Datasets. The draft pseudo-topic is never offered.

  • Sorts. Every sort ends with title ascending, then pk.

    Key Column Expression Nulls
    dataset Dataset lower-cased title, falling back to the name –
    status Status published_at drafts lowest
    tables Tables member count as a subquery Count, never a join –
    modified Modified modified_at last both ways
    created Created created_at –

    Default -modified.

  • 25 per page. The tables tab's URL rules hold:

    • defaults are never written; the bare address is unfiltered and -modified;
    • any filter change goes to page 1;
    • unknown values are ignored and shown as chips, never refused;
    • multi-values are comma-joined.
  • The old datasets_page parameter is dropped without a redirect. Datasets are not used on
    production yet.

  • The base queryset is the user's own Datasets: no join, no distinct().

  • Query budget per htmx list request: 6 unfiltered, 8 with Topic and Tag both named.

    • The 6: session, user, the faceted segment aggregate, the page (with member count and the
      popover's draft and embargoed counts as subquery annotations), member titles, the Topics
      prefetch.
    • Member titles are capped in SQL at ten per Dataset with one windowed query, not a whole
      prefetch.
  • The filter bar: Search · Topic · Tag · "More filters (n)" · New dataset. Narrow: everything
    but Search folds behind "Filters (n)".

  • The empty states:

    • No Datasets: an explanation and "Create a dataset", with no bar, segment or header.
    • No match: "No datasets match these filters" and Reset filters, header kept.
Page layout and narrow screens (measured on the prototype)
  • The datasets tab keeps the profile sidebar, like the tables tab. The profile keeps opening on
    the datasets tab
    (maintainer). Tab order: Datasets · Tables · ….

  • Collapse keys on the list's own width (container queries, scoped to this tab). These were
    measured in headless Chrome with DejaVu Sans, against the real tables tab's frame:

    List width below Change
    920 px Created hidden
    810 px Topics hidden too
    660 px rows stacked
    630 px filter bar folded
    • With the sidebar: every column at 1920, 1440, 1366 and 1280; without Created and Topics at 1024;
      stacked at 800 and 390.
    • The filter bar folds only where the rows already stack.
    • The first slice that renders the real row re-measures these with the parameterised width
      script and writes them beside the queries.
      Expect movement of tens of pixels.
  • Stacked row: line 1 is ☐, title and ⋯; line 2 is Status, Tables and Modified. A draft's
    missing-Topic hint is not shown there (accepted). "Sort by" and "Select this page" replace the
    headers.

Dashboard endpoints (thin, owner rule on every one)

Everything that is not one Dataset sits beside datasets/, never under it, because a Dataset
name may be actions or names.

Route Purpose
profile/<id>/datasets the list page; the results region on htmx
profile/<id>/dataset-actions/<action> GET preflight/dialog, POST execute: create, edit, publish, unpublish, delete
profile/<id>/dataset-actions/<action>/check the bulk bar's POST preflight
profile/<id>/dataset-names "Select all N matching" over the current filters
profile/<id>/dataset-name-preview the live name preview
profile/<id>/datasets/<name>/members the drawer (GET); add and remove (POST)
profile/<id>/datasets/<name>/members/search the drawer's add search
  • Events: datasets-changed (the list re-fetches; stay: true from the drawer),
    datasets-refused and datasets-results. They go through the generic action-view base with the
    event names as attributes.
  • A selection travels as one comma-joined field: Django refuses more than 1,000 parameters.
  • htmx contract as on the tables tab: HX-Push-Url carries the canonical address, and a history
    restore request is answered with the full page.
  • Every new route joins the owner-rule test's route lists. Not the user's own Dataset → 404.
Members drawer
  • An offcanvas outside the list region, opened by "Manage tables…", by the popover's "and n more",
    and after Create.
  • Members section: every member, 25 per page, a search over members, Draft and Embargoed badges,
    a neutral Remove per member.
    • Remove acts at once, except for a member the user could no longer add back (a draft or
      embargoed Table they hold no Data editor role on). That asks first: "You will not be able to add
      it back while it is a draft / embargoed."
      This is read with the same assignment rule, one query
      per page.
    • Removing the last member of a published Dataset is allowed; the drawer says it stays published.
  • Add section:
    • a search over assignable_tables(user), by title and name, paged, not capped;
    • before anything is typed it lists the user's own assignable Tables (maintainer);
    • an Add button per result turns into "Added"; existing members read "Already in";
    • when adding seeded Topics, the message names them;
    • the drawer's gate line reads "choose at least one topic (Edit…, or add a table that has one)".
  • The hand-off link "Work on your own tables in this dataset in the tables tab →" goes to the
    tables tab filtered to this Dataset. It is shown only when the user holds at least one member, with
    the count of members that are theirs.
  • The drawer is in the list's address as ?members=<name> (maintainer), so browser Back from the
    hand-off reopens it and the drawer can be linked.
    • It is page state, not list state: the tab's script writes it with replaceState on open and
      removes it on close, so it adds no history entry.
    • On load and on history restore, the script opens the drawer when it names one of the user's
      Datasets; any other value is ignored.
    • The canonical list address, the pager, sort and filter links never carry it.
  • Each change answers with the re-rendered drawer and datasets-changed with stay: true, so the
    list refreshes behind it without moving focus.
  • No multi-select in the drawer in v1.
Create and Edit
  • One modal form: title, description, Topics (optional on create, with a note that one is needed
    to publish).

  • The derived name previews live as the title is typed, after a 300 ms pause. The four states:

    • "Web address: …/datasets/‹name›. Fixed once created.";
    • "already taken";
    • "held" (with the held-name sentence);
    • "needs a letter or number".

    Create is enabled only when the name is available. Edit shows the name read-only.

  • A refused save re-renders in the dialog with what was typed. Success: the dialog closes,
    datasets-changed fires, a toast confirms.

  • After Create, in this order: the dialog closes → the list re-fetches (the new draft lands on
    top) → the members drawer opens with focus in the add search.

  • The first Create in an empty account must also bring in the filter bar and the bulk slot,
    which sit outside the swapped region (out-of-band swap). Deleting the last Dataset does the
    reverse.

Row actions, bulk bar, dialogs
  • ⋯ menu: Edit… · Manage tables… · Publish… or Unpublish… · divider · Delete… (red, last).
    Every entry is enabled on every row.
  • Publish on a draft failing the gate opens a dialog that names what is missing, with
    "Manage tables…" and "Edit…" links and no confirm button. The links close the dialog first and
    open their target once it has closed
    , because a modal cannot open a drawer or another modal while
    it is showing.
  • There is no republish in the dashboard; the API keeps it.
  • The publish dialog: where the Dataset will be listed, plus the member mix as information.
  • The unpublish dialog: the consequences in words, no OEKG link count.
  • The delete dialog: what was decided above, no OEKG link count. The hold is silent at delete.
  • Bulk bar (sticky, in a reserved slot showing a muted line when nothing is selected):
    Publish…, Unpublish…, Delete…, collapsing into an "Actions" menu where the rows stack.
    • Left-out groups have singular and plural forms.
    • Bulk publish leaves out gate failures, with the failed check, and already published Datasets.
    • Bulk unpublish leaves out drafts.
    • Bulk delete uses the typed count rule.
  • Selection: an explicit name list, "Select all N matching" through dataset-names, cleared on
    a filter change, pruned of Datasets that left the list.
  • No bulk Edit and no bulk member change.
  • Inherited from the tables tab without change:
    • one re-fetch after every action;
    • one toast region (success polite for about 5 s; refusals assertive and persistent; 401 → "You
      have been logged out" plus a login link with next; 5xx → "nothing may have changed");
    • one global htmx error handler;
    • a row that left the view is named;
    • focus returns to the row's ⋯ or the list heading;
    • a refused batch keeps the dialog, says "Nothing was changed" and re-runs the preflight.
Accessibility (inherited, final)
  • Native tab order.
  • Stable ids on every control in the region, with focus restored by id after a swap.
  • The aria-live result count and the bulk bar live outside the swapped region.
  • Popovers with links take focus and give it back on Escape.
  • "–" cells use visually hidden text, never aria-label on a span.
  • Anything ever disabled uses aria-disabled, never Bootstrap's .disabled.
  • Screen-reader labels: "Select ‹title›", "Actions for ‹title›", "5 tables, 2 drafts. Show them",
    "No tables. At least one is needed to publish."
The Dataset REST API
Capability Route
list (visibility as above) GET /datasets/ with ?mine=true, ?published=true|false, page, page_size
create POST /datasets/ (incl. topics)
read GET /datasets/<name>/
edit PATCH /datasets/<name>/ (title, description, at_id, topics)
publish / republish POST /datasets/<name>/publish/
unpublish POST /datasets/<name>/unpublish/
delete DELETE /datasets/<name>/
members GET …/resources/ (paginated), POST …/assign-tables/, POST …/unassign-tables/

Read body

  • It carries published_at (null = draft; no derived state field), creator (the username, null
    when ownerless), topics (sorted names), created_at and modified_at. All are read-only.
  • List items are summaries: the detail body without metadata.resources, plus resource_count.
  • The list uses page / page_size, default 20, maximum 100, in DRF's {count, next, previous, results} envelope, and so do the resources.
  • Constant queries per page.

Writes

  • Topics: optional on create (omitted = none). On PATCH: omitted = untouched, a list replaces
    the set, [] empties it. Unknown names and the draft pseudo-topic are a 400 naming them.
  • PATCH replaces PUT, which answers 405. A missing key is untouched. A name in the body is
    a 400, even when equal.
    It goes through the shared update path, so an omitted at_id keeps the
    stored one.
  • Publish and unpublish: no body, through the Dataset action service with via="api".
    • Success is 200 with the read body.
    • A missed gate is 409 {"detail": "…", "failed": ["members", "topics"]}.
    • Unpublishing a draft is a 200 that writes nothing; publishing a published Dataset republishes it.
  • Delete: through the service. The URL's name is the typed confirmation. It answers a bare
    204; a repeat is 404.
  • Member routes: bodies and the missing semantics are unchanged. Assign refuses the whole
    request with a 403 naming the Tables outside the assignment rule. The routes resolve through the
    read rule first, and now go through members_add / members_remove with via="api". The new
    2,500 ceiling
    answers 400 naming it.

Refusals and lookups

  • Bodies speak DRF throughout: {"detail": …}, or the field map for validation. The
    {"error": …} shape goes. A foreign draft and an unknown name return the same 404 body, word for
    word.
  • The ?mine=true and ?published= filters: anonymous mine=true → 401; unknown values → 400
    naming the parameter.
  • The resources read declares its permission classes instead of inheriting DRF's default.

The OpenAPI description (following #2480)

  • Every operation declares its success body from read serializers, and its refusals with one
    sentence each.
  • Every description states its auth expectation and the draft rule in prose: "Public. A draft is
    visible only to its creator; anyone else gets 404."
  • The datasets tag description gains a lifecycle sentence.
  • openapi.yaml is regenerated.

Parity is capability, not batch shape. Every dashboard action has a route above. A bulk
dashboard action maps to the single route called once per Dataset. There is no bulk API route.

OEKG
  • Resolution of a ref: dataset link reads the lifecycle: a draft target resolves as
    "unknowable" (resolvable: null, tables: null) for every reader, the creator included. A deleted
    target reads resolvable: false, as today.
  • No change to how links are written.
Transition
  • The old card routes (card, edit, delete, manage, table-search, assign,
    unassign), their partials and the dataset_creator_required decorator are retired in the same
    slice that brings the new routes
    . Its 403 becomes the owner rule's 404.
  • Tests that pin today's shape change:
    • "every Dataset is public";
    • the card, manage, assign and unassign view tests;
    • every API PUT test;
    • the 204-with-body assertion;
    • the four tables-tab dialog tests naming other people's Datasets;
    • the OEKG Dataset resolution test.
  • Fixtures that create a Dataset with a bare create become drafts, and set published_at where
    they need a published one.
  • Deploy needs: the Django migrations (the two fields with their backfill, and the hold table),
    and the regenerated openapi.yaml. No OEDB migration. These go into the next release's deploy
    checklist.

Testing Decisions

What makes a good test here: it exercises behaviour through the highest seam a user or client
touches, and asserts on structure, never on markup detail or seconds:

  • rows and their order;
  • faceted counts and stale chips;
  • which Datasets a preflight leaves out and why;
  • the database state after a write, and a refused request writing nothing;
  • response codes, bodies and HX-Trigger headers;
  • the log lines;
  • query counts.

The seams (confirmed by the maintainer):

  1. HTTP through Django's test client.
    • The dashboard: login:datasets and each new login/ endpoint, including the region on htmx,
      each action's GET and POST, the bulk check, dataset-names, the name preview, and the drawer
      with its search and writes.
    • The Dataset REST API: every route above.
    • The public readers: the topic list, the detail page, the metadata JSON, and the table page's
      Datasets sidebar.
    • The tables tab's existing Datasets column, filter and dialogs, under the new visibility.
    • Prior art: the tables-tab suites (test_tables_tab, test_table_actions, test_tables_bulk,
      test_table_access), the structural owner-rule test, api/tests/test_datasets_api.py.
    • Query bounds with assertNumQueries:
      • the list's 6 / 8, at 3 and 30 Datasets and with one Dataset of 2,500 members, unchanged by
        any of them;
      • the API list and resources pages constant per page;
      • the drawer's remove-confirm read, one query per page.
    • One test compares a foreign draft's 404 body with an unknown name's, word for word.
    • Every new route joins the owner-rule route lists.
  2. The generic list module under vitest + happy-dom. Prior art:
    login/static/login/__tests__/tables_tab.test.js, which must run unchanged through the refactor.
    • Selection across paging and sorting, cleared on a filter change with a note, pruned of names
      that left the list.
    • "Select all N matching", Shift ranges, the reserved bulk slot.
    • The ?members= handling: written on open, removed on close, the drawer reopened on load.
  3. A response-schema test: one real response of every Dataset API operation, success and each
    declared refusal, validated against the committed openapi.yaml. Prior art:
    oekg/tests/test_response_schema.py, whose JSON-schema translation helper moves to a shared test
    module rather than being copied. The existing OpenAPI drift guard stays.
  4. call_command for drop_dead_dataset_links against a real Fuseki, through the OEKG graph test
    case, which skips cleanly without a store. The dry run changes nothing; --apply removes the links
    and the hold; the record is written before the graph.
    • Plus: the delete writes a hold when the ASK finds a citation and when the store is
      unreachable, and frees the name when it does not.

Also:

  • The migration's backfill is guarded structurally, as modelview's delete-grace migration test
    does: existing rows get published_at = created_at and modified_at null. Run sqlmigrate to
    check that no DEFAULT now() lands on existing rows (the auto_now_add trap of modelview.0066
    and dataedit.0057).
  • Not covered by any seam: the collapse thresholds. happy-dom does no layout. They are measured
    once in a real browser with the width script and written beside the container queries. There is no
    browser end-to-end suite.
  • The bulk ceiling for publish, unpublish and delete is measured with a benchmark beside the
    tables tab's, in the slice that adds them.

Out of Scope

  • Dataset roles: co-editors, Organizations, an access drawer for Datasets. They belong to the
    separate roles effort. The column list allows an Access column later without a rebuild.
  • Other people's Datasets that contain my Tables, as a list or segment on this tab. The tables
    tab's Datasets column and its Dataset filter are that view.
  • Adopting ownerless Datasets, and what happens to a leaving user's drafts. This waits on
    account deletion (#1181) and the roles effort.
  • Redesigning the public dataset pages beyond hiding drafts.
  • The full fix for links to a deleted Dataset: telling the citing bundles' owners, their choice
    during the grace period, a scheduled cleanup. drop_dead_dataset_links stands in until then. The
    same question for deleted Table names
    (re-takable while ref: table resolves by name) is a
    platform question.
  • A Dataset embargo, and a stored publishable flag.
  • API filters beyond mine and published (creator, search, Topic, dates), and a bulk API route.
    All of these can be added later.
  • Platform-wide unification of error bodies (reason / error / detail) outside the Dataset API.
  • Multi-select in the members drawer.
  • A browser end-to-end suite.

Further Notes

  • Not in v1.11.0 (freeze 2026-10-07). Until this ships, every Dataset stays publicly listed. That
    is harmless while production holds only test Datasets.
  • oemetadata publicationDate is a different fact from published_at: when the data was
    published, possibly long before the OEP, versus when the Dataset became public here. The next
    oemetadata version moves it to the Dataset level; the sync rule is decided with that upgrade. A
    starting point: on publish, fill it if empty, never overwrite.
  • CONTEXT.md's Dataset link entry says a link "is never cleaned up". The cleanup command
    makes that untrue for Dataset links cited after a delete. The entry is amended with the slice that
    ships the command.
  • Open for review, not decided: whether the members drawer should mark which members are the
    user's own. The hand-off count and the remove confirmation carry that difference today.
  • The tables tab's version of the route collision (tables/actions/…) is #2611, fixed in PR
    #2612.
  • The prototype, its measurements and everything the build found are in the map's assets
    (WF-07 - prototype notes), in the maintainer's vault.
  • No ADR: every decision here is reversible UI or service structure. The one departure from a
    closed decision, the OEKG link cleanup, is recorded on that decision itself.

🤖 Generated with Claude Code

Lingua principale
Python
Stelle
65
Fork
28
Merge medio
4g 7h
PR unite (30g)
25

Preparare l'ambiente

Apri in Codespaces

Avvia il container di sviluppo del progetto nel browser, con il tuo account GitHub.

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di OpenEnergyPlatform/oeplatform

Tutte le issue di OpenEnergyPlatform/oeplatform

Issue simili

Altre issue su Python

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.