Summary:
Ref T13552. Give "Commit" objects a more modern, identity-aware way to render author and committer information.
This uses handles in a more modern way and gives us a single read callsite for raw author and committer names.
Test Plan:
- Grepped for callers to the old methods, found none. (There are a lot of "renderAuthor()" callers in transactions, but this call takes no parameters.)
- Viewed some commits, saw sensible lists of authors and committers.
Maniphest Tasks: T13552
Differential Revision: https://secure.phabricator.com/D21405
Summary:
Ref T13552. When viewing a directory in Diffusion, we make an Ajax call to get the last commit for each path.
This call currently pulls author information, since an older version of this UI showed author information.
The current UI does not show author information, so this parameter is unused. Delete the code which builds it.
Test Plan: Grepped for `'author'` and references to the "pull-lastmodified" behavior. This behavior is invoked in only one place, which never generates an author placeholder.
Maniphest Tasks: T13552
Differential Revision: https://secure.phabricator.com/D21404
Summary:
Modern Mercurial may emit some more patterns under "--debug".
This whole list is gross and can likely now be eliminated by increasing the minimum required Mercurial version (as `arc` has), but just paper over it for now.
Test Plan:
Locally, saw some views return to functional behavior that weren't previously working on a modern version of Mercurial.
The reproduction case is likely something in the vein of "repository is not writable by webserver, look at history view".
Differential Revision: https://secure.phabricator.com/D21398
Summary:
Ref T13541. The passthru future does not have time limit behavior, so if we reach this code we currently fail.
Phabricator never reaches this code normally, but this code is reachable during debugging if you try to foreground a slow fetch to inspect it.
Passthru commands generally only make sense to run interactively, and the caller or control script can enforce their own timeouts (usually by pressing "^C" with their fingers).
Test Plan: Used a debugging script to run ref-by-ref fetches in the foreground.
Maniphest Tasks: T13541
Differential Revision: https://secure.phabricator.com/D21284
Summary: Ref T13513. Inline comment context information is somewhat expensive to construct and can be cached. Add a readthrough cache on top of it.
Test Plan: Loaded a source code changeset with many inline comments, used Darkconsole to inspect query activity. Saw caches get populated. Updated cache key, saw caches regenerate. Browsed Diffusion, nothing looked broken.
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21279
Summary: Ref T13513. For now, I'm not supporting inline edit suggestions in Diffusion, although it's likely not difficult to do so in the future. Clean up some of the code so that plain ol' inlines work the same way they did before.
Test Plan:
- Created, edited, reloaded, submitted inlines in Diffusion: familiar old behavior.
- Created, edited, reloaded, submitted inlines with suggestions in Differential: familiar new behavior.
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21278
Summary:
Ref T13513. If your 10 most recently authored inlines have all been deleted, these queries can fail by overheating. This is silly and probably rarely happens outside of development.
For now, just let them overheat. This may create a false negative (incorrect "no draft" signal when the real condition is "drafts, but 10 most recent comments were deleted"). This could be sorted out later with a query mode like "executeAny()", perhaps.
Test Plan:
- Created and deleted 10 inlines.
- Submitted comments.
- Before: overheating fatal during draft flag generation.
- After: clean submission.
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21274
Summary:
Ref T13513. Currently, if you:
- click a line to create an inline;
- type some text;
- wait a moment; and
- close the page.
...you don't get an "Unsubmitted Draft" marker in the revision list.
Lift all the draft behavior to "InlineController" and make saving a draft dirty the overall container draft state.
Test Plan:
- Took the steps described above, got a draft state marker.
- Created, edited, submitted, etc., inlines in Diffusion and Differential.
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21235
Summary: Ref T13513. All queries now go through a reasonably minimal set of pathways and should have consistent behavior.
Test Plan:
- Loaded a revision with inlines.
- Created a new empty inline, reloaded page, saw it vanish.
- Created a new empty inline, typed draft text, did not save, reloaded page, saw draft present.
- Created a new empty inline, typed draft text. Submitted feedback, got prompt, answered "Y", saw draft text submit.
- Created a new empty inline, typed draft text, scrolled down to bottom of page, typed non-draft text, saw preview include draft text.
- Marked and submitted "Done".
- Used hide/show on inlines, verified state persisted.
- Did much of the same stuff in Diffusion, where it all works the same way (except: there's no prompt when submitting draft is-editing inlines).
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21234
Summary: Ref T13513. Replaces "DifferentialInlineCommentQuery" with the similar but more modern "DifferentialDiffInlineCommentQuery".
Test Plan: Viewed comments in timeline, changesets. Created, edited, and submitted comments. Hid and un-hid comments, reloading (saw state preserved).
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21233
Summary: Ref T13513. Continue removing usage sites for the obsolete "DifferentialInlineCommentQuery".
Test Plan: Viewed the inline list in Differential, saw sensible inlines.
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21232
Summary: Ref T13513. Move querying to "DiffInlineCommentQuery" classes and lift them into the base Controller.
Test Plan: In Differential and Diffusion, created, edited, and submitted inline comments.
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21231
Summary: Ref T13513. Another step closer to the light.
Test Plan: Created, edited, deleted, replied to, and submitted inline comments in Diffusion.
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21230
Summary: Ref T13513. Continue marching toward coherent query pathways for all access to inline comments.
Test Plan:
- Viewed a commit and a path within that commit, as a user with unpublished inlines and a different user.
- Saw appropriate inlines in all cases (published inlines, plus undeleted unpublished inlines authored by the current viewer).
- Grepped for "loadDraftAndPublishedComments()".
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21228
Summary:
Ref T13513. Improve consistency and robustness of the "InlineComment" queries.
The only real change here is that these queries now implicitly add a clause for selecting inlines ("pathID IS NULL" or "changesetID IS NULL").
Test Plan: Browed, created, edited, and submitted inlines.
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21227
Summary: Ref T13513. This controller was obsoleted by EditEngine and appears unreachable without explicitly typing the URL.
Test Plan:
- Grepped for the route, didn't find any hits.
- Deleted the controller, successfully previewed comments in Diffusion.
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21224
Summary:
Ref T13513. This is mostly an infrastructure cleanup change.
In a perfect world, this would be a series of several changes, but they're tightly interconnected and don't have an obvious clean, nontrivial partition (or, at least, I don't see one). Followup changes will exercise this code repeatedly and all of these individual mutations are "obviously good", so I'm not too worried about the breadth of this change.
---
Inline comments are stored as transaction comments in the `PhabricatorAuditTransactionComment` and `DifferentialTransactionComment` classes.
On top of these two storage classes sit `PhabricatorAuditInlineComment` and `DifferentialInlineComment`. Historically, these were an indirection layer over significantly different storage classes, but nowadays both storage classes look pretty similar and most of the logic is actually the same. Prior to this change, these two classes were about 80% copy/pastes of one another.
Part of the reason they're so copy/pastey is that they implement a parent `Interface`. They are the only classes which implement this interface, and the interface does not provide any correctness guarantees (the storage objects are not actually constrained by it).
To simplify this:
- Make `PhabricatorInlineCommentInterface` an abstract base class instead.
- Lift as much code out of the `Audit` and `Differential` subclasses as possible.
- Delete methods which no longer have callers, or have only trivial callers.
---
Inline comments have two `View` rendering classes, `DetailView` and `EditView`. They share very little code.
Partly, this is because `EditView` does not take an `$inline` object. Historically, it needed to be able to operate on inlines that did not have an ID yet, and even further back in history this was probably just an outgrowth of a simple `<form />`.
These classes can be significantly simplified by passing an `$inline` to the `EditView`, instead of individually setting all the properties on the `View` itself. This allows the `DetailView` and `EditView` classes to share a lot of code.
The `EditView` can not fully render its content. Move the content rendering code into the view.
---
Prior to this change, some operations need to work on inlines that don't have an inline ID yet (we assign an ID the first time you "Save" a comment). Since "editing" comments will now be saved, we can instead create a row immediately.
This means that all the inline code can always rely on having a valid ID to work with, even if that ID corresponds to an empty, draft, "isEditing" comment. This simplifies more code in `EditView` and allows the "create" and "reply" code to be merged in `PhabricatorInlineCommentController`.
---
Client-side inline events are currently handled through a mixture of `ChangesetList` listeners (good) and ad-hoc row-level listeners (less good). In particular, the "save", "cancel", and "undo" events are row-level. All other events are list-level.
Move all events to list-level. This is supported by all inlines now having an ID at all stages of their lifecycle.
This allows some of the client behavior to be simplified. It currently depends on binding complex ad-hoc dictionaries into event handlers in `_drawRows()`, but it seems like almost all of this code can be removed. In fact, no more than one row ever seems to be drawn, so this code can probably be simplified further.
---
Finally, save an "isEditing" state. When we rebuild a revision on the client, click the "edit" button if it's in this state. This is a little hacky, but simpler to get into a stable state, since the row layout of an inline depends on a "view row" followed by an "edit row".
Test Plan:
- Created comments on either side of a diff.
- Edited a comment, reloaded, saw edit stick.
- Saved comments, reloaded, saw save stick.
- Edited a comment, typed text, cancelled, "unedited" to get state back.
- Created a comment, typed text, cancelled, "unedited" to get state back.
- Deleted a comment, "undeleted" to get state back.
Weirdness / known issues:
- Drafts don't autosave yet.
- Fixed in D21187:
- When you create an empty comment then reload, you get an empty editor. This is a bit silly.
- "Cancel" does not save state, but should, once drafts autosave.
- Mostly fixed in D21188:
- "Editing" comments aren't handled specially by the overall submission flow.
- "Editing" comments submitted in that state try to edit themselves again on load, which doesn't work.
Subscribers: jmeador
Maniphest Tasks: T13513
Differential Revision: https://secure.phabricator.com/D21186
Summary:
Ref T13455. Make "hidden" a changeset property similar to other changeset properties.
We don't need to render this on the server, so we make a request (to update the setting) and just discard the response.
Test Plan: {F7375468}
Maniphest Tasks: T13455
Differential Revision: https://secure.phabricator.com/D21158
Summary:
Ref T13516. Deletes all old filetree / flex / active / collapse nav code in favor of the new code.
Restores the inline tips in the path tree.
Test Plan: {F7374175}
Maniphest Tasks: T13516
Differential Revision: https://secure.phabricator.com/D21154
Summary:
Ref T13515. This restores the "Open in Editor" behavior to Diffusion, and makes "\" work there.
The URI pattern is now sent as a structured template to the client, so the code will work properly if a file path contains "%l".
Test Plan:
- Clicked "Open in Editor" and pressed "\" in Diffusion when viewing a file.
- Clicked a line, hit "\", got the file opened to that line.
Maniphest Tasks: T13515
Differential Revision: https://secure.phabricator.com/D21149
Summary:
Ref T13515. Currently, opening a file to a particular line in an external editor relies on replacing "%l" with "%l" (which is escaped as "%25l") on the server, and then replacing "%25l" with the line number on the client. This will fail if the file path (or any other variable) contains "%l" in its unencoded form.
The parser also can't identify invalid variables.
Pull the parser out, formalize it, and make it generate an intermediate representation which can be sent to the client and reconstituted.
(This temporarily breaks Diffusion and permanently removes the weird, ancient integration in Dark Console.)
Test Plan:
- Added a bunch of tests for the actual parser.
- Used "Open in Editor" in Differential.
Maniphest Tasks: T13515
Differential Revision: https://secure.phabricator.com/D21143
Summary:
Ref T13455. Update the other "view state" properties to work like "highlight" now works.
Some complexity here arises from these concerns:
- In "View Standalone", we render the changeset inline. This is useful for debugging/development, and desirable to retain.
- In all other cases, we render the changeset with AJAX.
So the client needs to be able to learn about the "state" properties of the changeset on two different flows. Prior to this change, each pathway had a fair amount of unique code.
Then, some bookkeeping issues:
- At inital rendering time, we may not know which renderer will be selected: it may be based on the client viewport dimensions.
- Prior to this change, the client didn't separate "value of the property for the changeset as rendered" and "desired value of the property".
Test Plan:
- Viewed changes in Differential, Diffusion, and in standalone mode.
- Toggled renderer, character sets, and document engine (this one isn't terribly useful). Reloaded, saw them stick.
- Started typing a comment, cancelled it, hit the undo UI.
Maniphest Tasks: T13455
Differential Revision: https://secure.phabricator.com/D21138
Summary:
Ref T13455. Add container-level storage for persistent view state, and persist "Highlight As..." inside it.
The storage generates a "PhabricatorChangesetViewState" configuration object as an output.
When preferences are expressed on a diff and that diff is later attached to a revision, we attempt to copy the preferences.
The internal storage tracks per-changeset settings, but currently always uses "last update wins" to apply the settings in the UI.
Test Plan:
- Viewed revisions, changed highlighting, reloaded. Saw highlighting stick in revision view and standalone view.
- Viewed commits, changed highlighting, reloaded. Saw highlighting stick.
- Created a diff, changed highlighting, turned it into a revision, saw highlighting persist.
Subscribers: jmeador, PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13455
Differential Revision: https://secure.phabricator.com/D21137
Summary:
Fixes T5028. Older versions of Git (apparently, from before 2010) did not provide a way to extract the raw body of a commit message from "git log", so we approximate it with "subject" and "wrapped body".
In newer versions of Git, the raw body can be extracted exactly.
Adjust how we extract messages based on the version of Git, and try to be more faithful to edge cases: particularly, be more careful to extract the correct number of trailing newlines.
Test Plan:
- Added "var_dump()" + "die(1)" later in this method, then pushed various commit messages. Used "&& false" to force execution down the old path (either path should work in modern Git).
- Observed more faithful extraction of messages, including a more faithful extraction of the number of trailing newlines. Extraction is fully faithful if we can go down the "%B" path, which we should be able to in nearly all modern cases.
- Not all messages extract faithfully or consistently across the old and new versions, but the old extraction is destructive so this is likely about as close as we can realistically ever get.
Maniphest Tasks: T5028
Differential Revision: https://secure.phabricator.com/D21027
Summary:
`diffusion.branchquery` can return dictionary instead of array if some branches are filtered out.
Eg.:
```
{
"result": [
{
"shortName": "master",
"commitIdentifier": "2817b0d8f79748ddfad0220c46d9b20bea34f460",
"refType": "branch",
"rawFields": {
"objectname": "2817b0d8f79748ddfad0220c46d9b20bea34f460",
"objecttype": "commit",
```
might become:
```
{
"result": {
"1": {
"shortName": "master",
"commitIdentifier": "2817b0d8f79748ddfad0220c46d9b20bea34f460",
"refType": "branch",
"rawFields": {
"objectname": "2817b0d8f79748ddfad0220c46d9b20bea34f460",
"objecttype": "commit",
```
Reproduction - find repository which has couple of branches, setup to track only some of them, execute `diffusion.branchquery` API call - result is dictionary instead of array
Test Plan: Apply patch, execution `diffusion.branchquery` call - result is no longer dictionary if it was one before
Reviewers: #blessed_reviewers, epriestley
Reviewed By: #blessed_reviewers, epriestley
Subscribers: Korvin, epriestley
Differential Revision: https://secure.phabricator.com/D20973
Summary: Depends on D20931. Ref T13362. Move all "Console"-style interfaces to use a consistent layout based on a new "LauncherView" which just centers the content.
Test Plan: Viewed all affected interfaces.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13362
Differential Revision: https://secure.phabricator.com/D20933
Summary: Ref T13480. Some Herald fields need audit information, which recent changes to Herald adapters discarded. For now, just load it unconditionally.
Test Plan: Triggered an Audit-related rule locally.
Maniphest Tasks: T13480
Differential Revision: https://secure.phabricator.com/D20962
Summary: Fixes T13480. Adds the remaining missing Owners package rules for Herald commit adapters.
Test Plan: Created hooks which care about these fields, pushed commits, saw sensible transcript values.
Maniphest Tasks: T13480
Differential Revision: https://secure.phabricator.com/D20957
Summary: Ref T13480. The Herald "Commit" rules still use raw commit data properties to identify authors and committers. Instead, use repository identities.
Test Plan: Wrote a Herald rule using all four fields, ran it against various commits with and without known authors. Checked transcript for sensible field values.
Maniphest Tasks: T13480
Differential Revision: https://secure.phabricator.com/D20955
Summary:
Ref T13480. Currently, Herald commit hook rules use a raw address resolution query to identify the author and committer for a commit. This will get the wrong answer when the raw identity string has been explicitly bound to some non-default user (most often, it will fail to identify an author when one exists).
Instead, use the "IdentityEngine" to properly resolve identities.
Test Plan: Authored a commit as `X <y@example.com>`, a raw identity with no "natural" matches to users (e.g., no user with that email or username). Bound the identity to a particular user in Diffusion. Wrote a Herald pre-commit content rule, pushed the commit. Saw Herald recognize the correct user when evaluating rules.
Maniphest Tasks: T13480
Differential Revision: https://secure.phabricator.com/D20953
Summary: Ref T13444. Allow the effects of performing an identity rebuild to be previewed without committing to any changes.
Test Plan: Ran "bin/repository rebuild-identities --all-identities" with and without "--dry-run".
Maniphest Tasks: T13444
Differential Revision: https://secure.phabricator.com/D20922
Summary:
Ref T13444. Currently, many mutations to users and email addresses (particularly: user creation; and user and address destruction) do not propagate properly to repository identities.
Add hooks to all mutation workflows so repository identities get rebuilt properly when users are created, email addresses are removed, users or email addresses are destroyed, or email addresses are reassigned.
Test Plan:
- Added random email address to account, removed it.
- Added unassociated email address to account, saw identity update (and associate).
- Removed it, saw identity update (and disassociate).
- Registered an account with an unassociated email address, saw identity update (and associate).
- Destroyed the account, saw identity update (and disassociate).
- Added address X to account A, unverified.
- Invited address X.
- Clicked invite link as account B.
- Confirmed desire to steal address.
- Saw identity update and reassociate.
Maniphest Tasks: T13444
Differential Revision: https://secure.phabricator.com/D20914
Summary:
Ref T13444. Repository identities have, at a minimum, some bugs where they do not update relationships properly after many types of email address changes.
It is currently very difficult to fix this once the damage is done since there's no good way to inspect or rebuild them.
Take some steps toward improving observability and providing repair tools: allow `bin/repository rebuild-identities` to effect more repairs and operate on identities more surgically.
Test Plan: Ran `bin/repository rebuild-identities` with all new flags, saw what looked like reasonable rebuilds occur.
Maniphest Tasks: T13444
Differential Revision: https://secure.phabricator.com/D20911
Summary: Ref T13444. Send all repository identity/detection through a new "DiffusionRepositoryIdentityEngine" which handles resolution and detection updates in one place.
Test Plan:
- Ran `bin/repository reparse --message ...`, saw author/committer identity updates.
- Added "goose@example.com" to my email addresses, ran daemons, saw the identity relationship get picked up.
- Ran `bin/repository rebuild-identities ...`, saw sensible rebuilds.
Maniphest Tasks: T13444
Differential Revision: https://secure.phabricator.com/D20910
Summary: Ref T13444. This is an ancient event and part of the old event system. It is not likely to be in use anymore, and repository identities should generally replace it nowadays anyway.
Test Plan: Grepped for constant and related methods, no longer found any hits.
Maniphest Tasks: T13444
Differential Revision: https://secure.phabricator.com/D20909
Summary:
Ref T13444. You can currently explicitly unassign an identity (useful if the matching algorithm is misfiring). However, this populates the "currentEffectiveUserPHID" with the "unassigned()" token, which mostly makes things more difficult.
When an identity is explicitly unassigned, convert that into an explicit `null` in the effective user PHID.
Then, realign "assigned" / "effective" language a bit. Previously, `withAssigneePHIDs(...)` actualy queried effective users, which was misleading. Finally, bulk up the list view a little bit to make testing slightly easier.
Test Plan:
- Unassigned an identity, ran migration, saw `currentEffectiveUserPHID` become `NULL` for the identity.
- Unassigned a fresh identity, saw NULL.
- Queried for various identities under the modified constraints.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13444
Differential Revision: https://secure.phabricator.com/D20908
Summary:
Ref T13444. Currently, identities for a particular email address are queried with "LIKE" against a binary column, which makes the query case-sensitive.
- Extract the email address into a separate "sort255" column.
- Add a key for it.
- Make the query a standard "IN (%Ls)" query.
- Deal with weird cases where an email address is 10000 bytes long or full of binary junk.
Test Plan:
- Ran migration, inspected database for general sanity.
- Ran query script in T13444, saw it return the same hits for "git@" and "GIT@".
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13444
Differential Revision: https://secure.phabricator.com/D20907
Summary: Ref T13448. Minor UI issue: setting a "Fetch Refs" value does not highlight the associated menu item, but should.
Test Plan: Set only "Fetch Refs", now saw menu item highlighted.
Maniphest Tasks: T13448
Differential Revision: https://secure.phabricator.com/D20895
Summary:
Ref T12164. Ref T13439. Commit hovercards don't currently show the repository. Although this is sometimes obvious from context, it isn't at other times and it's clearly useful/important.
Also, use identities to render author/committer information and show committer if the committer differs from the author.
Test Plan: {F6989595}
Maniphest Tasks: T13439, T12164
Differential Revision: https://secure.phabricator.com/D20881
Summary: Fixes T13430. Provide more information about repositories in "diffusion.repository.search".
Test Plan: Used API console to call method (with new "metrics" attachment), reviewed output. Saw new fields returned.
Maniphest Tasks: T13430
Differential Revision: https://secure.phabricator.com/D20862
Summary:
Fixes T13284. See that task for substantial discussion. There are currently two cases where we'll skip over commits which we should publish:
- if a branch is not permanent, then later made permanent; or
- in some cases, the first time we examine branches in a repository.
In both cases, this error is one-shot and things work correctly going forward. The root cause is conflation between the states "this ref currently permanent" and "this ref was permanent the last time we updated refs".
Separate these pieces of state and cover all these cases. Also introduce a "--rebuild" flag to fix the state of bad commits.
Test Plan:
See T13284 for the three major cases:
- initial import;
- push changes to a nonpermanent branch, update, then make it permanent;
- push chanegs to a nonpermanent branch, update, push more changes, then make it permanent.
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13284
Differential Revision: https://secure.phabricator.com/D20829
Summary:
Ref T13286. The current (very safe / conservative) rules for retrying git reads generalize to git writes, so we can use the same ruleset in both cases.
Normally, writes converge rapidly to only having good nodes at the head of the list, so this has less impact than the similar change to reads, but it generally improves consistency and allows us to assert that writes which can be served will be served.
Test Plan:
- In a cluster with an up node and a down node, pushed changes.
- Saw a push to the down node fail, retry, and succeed.
- Did some pulls, saw appropriate retries and success.
- Note that once one write goes through, the node which received the write always ends up at the head of the writable list, so nodes need to be explicitly thawed to reproduce the failure/retry behavior.
Maniphest Tasks: T13286
Differential Revision: https://secure.phabricator.com/D20778
Summary: Ref T13286. When retrying a read request, keep retrying as long as we have canididate services. Since we consume a service with each attempt, there's no real reason to abort early, and trying every service allows reads to always succeed even if (for example) 8 nodes of a 16-node cluster are dead because of a severed network link between datacenters.
Test Plan: Ran `git pull` in a clustered repository with an up node and a down node; saw retry count dynamically adjust to available node count.
Maniphest Tasks: T13286
Differential Revision: https://secure.phabricator.com/D20777
Summary:
Depends on D20775. Ref T13286. When a Git read request fails against a cluster and there are other nodes we could safely try, try more nodes.
We DO NOT retry the request if:
- the client read anything;
- the client wrote anything;
- or we've already retried several times.
Although //some// requests where bytes went over the wire in either direction may be safe to retry, they're rare in practice under Git, and we'd need to puzzle out what state we can safely emit.
Since most types of failure result in an outright connection failure and this catches all of them, it's likely to almost always be sufficient in practice.
Test Plan:
- Started a cluster with one up node and one down node, pulled it.
- Half the time, hit the up node and got a clean pull.
- Half the time, hit the down node and got a connection failure followed by a retry and a clean pull.
- Forced `$err = 1` so even successful attempts would retry.
- On hitting the up node, got a "failure" and a decline to retry (bytes already written).
- On hitting the down node, got a failure and a real retry.
- (Note that, in both cases, "git pull" exits "0" after the valid wire transaction takes place, even though the remote exited non-zero. If the server gave Git everything it asked for, it doesn't seem to care if the server then exited with an error code.)
Maniphest Tasks: T13286
Differential Revision: https://secure.phabricator.com/D20776
Summary:
Ref T13286. To support request retries, allow the service lookup method to return an ordered list of structured service references.
Existing callsites continue to immediately discard all but the first reference and pull a URI out of it.
Test Plan: Ran `git pull` in a clustered repository with an "up" node and a "down" node, saw 50% serivce failures and 50% clean pulls.
Maniphest Tasks: T13286
Differential Revision: https://secure.phabricator.com/D20775
Summary:
Ref T13393. While doing a shard migration in the Phacility cluster, we'd like to stop writes to the migrating repository. It's safe to continue serving reads.
Add a simple maintenance mode for making repositories completely read-only during maintenance.
Test Plan: Put a repository into read-only mode, tried to write via HTTP + SSH. Viewed web UI. Took it back out of maintenance mode.
Maniphest Tasks: T13393
Differential Revision: https://secure.phabricator.com/D20748
Summary:
Ref T13339. If a search pattern matches more than once on a line, we currently render the line incorreclty, duplicating some of the text.
`substr()` is being called as though the third parameter was `end_offset`, but it's actually `length`. Correct the parameter.
Test Plan:
Before:
{F6676625}
After:
{F6676623}
Maniphest Tasks: T13339
Differential Revision: https://secure.phabricator.com/D20695
Summary:
Fixes T8830. Fixes T13364.
- The inability to destroy objects from the web UI is intentional. Make this clear in the messaging, which is somewhat out of date and partly reflects an earlier era when things could be destroyed.
- `bin/remove destroy` can't rewind time. Document expectations around the "put the cat back in the bag" use case.
Test Plan: Read documentation, clicked through both workflows.
Maniphest Tasks: T13364, T8830
Differential Revision: https://secure.phabricator.com/D20694
