Summary:
Ref T13151. See T11767. See PHI686. Although we limit outbound mail text bodies, the limit doesn't currently apply to attachments, HTML bodies, or headers. T11767 discusses improving this in the general case.
In the wild, an install hit an issue (see PHI686) where edits to Phriction pages generate very large HTML bodies. Check and respect the limit when building HTML bodies.
If we don't have enough room for the HTML body, we just drop it. We have the text body to fall back to, and HTML is difficult to truncate safely.
Test Plan: Added unit tests and made them pass.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13151
Differential Revision: https://secure.phabricator.com/D19489
Summary:
Ref T13151. See PHI683. Ref T12314.
You can currently change object subtypes via Conduit (`maniphest.edit`) but not via the web UI.
Changing object subtypes is inherently a somewhat-perilous operation that likely has a lot of rough edges we'll need to smooth over eventually, mostly around changing an object from subtype X to subtype Y, where some field exists on one but not the other. This isn't a huge issue, just not entirely intuitive.
It should also, in theory, be fairly rare.
As a reasonable middle ground, provide web UI access via the bulk editor. This makes it possible, but doesn't clutter the UI up with a rarely-used option with rough edges.
Test Plan:
- With subtypes not configured, saw a normal bulk editor with no new option.
- With subtypes configured, swapped tasks subtypes via bulk editor.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13151, T12314
Differential Revision: https://secure.phabricator.com/D19490
Summary:
Depends on D19487. Ref T13151. See PHI647. For some objects, like revisions, we can build slightly more useful secure email without actually disclosing anything.
In the general case, the object monogram may disclose information (`#acquire-competitor`) but most do not, so applications can whitelist an acceptable nondisclosing subject and link.
Support doing this, and make Differential do it. When we don't have a whitelisted URI but do know the object the mail is about, include a generic PHID-based URI; these are always nondisclosing.
Test Plan:
- Without the Differential changes, sent normal mail (no changes) and secure mail (new generic PHID-based link).
- With the Differential changes, sent secure mail; got richer subject and body link.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13151
Differential Revision: https://secure.phabricator.com/D19488
Summary:
Ref T13151. See PHI647. This allows us to link to any object by PHID, without disclosing information in the monogram (like `#fire-steve`).
This capability is relevant when building "secure mail", to provide a link to the object regardless of whether the monogram discloses information or not.
Test Plan: Visited `/object/D123/` (redirect), `/object/xyz/` (404), `/object/PHID-DREV-.../` (redirect).
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13151
Differential Revision: https://secure.phabricator.com/D19487
Summary:
Ref T13151. See PHI702. An install is interested in a "members of all projects" (vs "members of any project", which is currently implemented) rule.
Although this is fairly niche, I think it's reasonable and doesn't have much of a maintenance cost.
This could already be implemented as an extension, but it would have to copy/paste a bunch of code.
Test Plan:
- Ran unit tests.
- Used the UI to select this policy for a task, with various values. Joined/left projects to satisfy/fail the rule. Behavior seemed correct.
- Used the UI to select the existing policy rule ("any project"), joined/left projects to satisfy/fail the rule. Doesn't look like I broke anything.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13151
Differential Revision: https://secure.phabricator.com/D19486
Summary:
Ref T13151. See PHI701. Unified diffs are currently missing the logic to apply the "old-full" and "new-full" classes, which results in a too-light coloration for fully added or removed lines.
Make this logic consistent with the two-up renderer so we use the same colors in both.
Test Plan: Viewed diffs and swapped between 1-up and 2-up renderers, now saw the same coloration.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13151
Differential Revision: https://secure.phabricator.com/D19482
Summary:
Ref T13151. See PHI685. When you haunt the panel, we only let it take up part of the screen. Let it take up slightly more of the screen so that it's more likely to fit completely on-screen without needing to scroll.
The behavior when it does scroll is fine (you get a scrollbar if your OS/browser is set up to show them) so this is a bit trivial/silly, but seems fine and doesn't have a big JS maintenance cost or anything.
Test Plan: Pressed "Z", resized my window to a weird tiny useless size, got slightly better (I guess) behavior.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13151
Differential Revision: https://secure.phabricator.com/D19480
Summary: Ref T13152. The pager does a bit of magic here and doesn't populate `nextPageID` when it knows it got an exact final page. The logic misfired in this case and sent us back to the start.
Test Plan:
- Set page size to 1 to guarantee rows were an exact multiple of page size.
- Ran `rebuild-identities` (I no-op'd the actual logic to make it faster).
- Before: looped forever.
- After: clean exit after processing everything.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13152
Differential Revision: https://secure.phabricator.com/D19479
Summary:
Ref T13151. See PHI654. Depends on D19477. If you have long package names, the table of contents (e.g., in Differential) can end up expanding to be gigantic.
Getting tables to behave nicely is hard (or, at least, I can't figure it out after spending a decent amount of time on it; see also `AphrontTableView::renderSingleDisplayLine()`). I tried a bunch of things and Googled for a bit but didn't make any progress on finding a CSS solution. Just truncate the package names to get reasonable behavior without falling down any kind of CSS rabbit hole.
Test Plan:
- Created a package named "Very long package name...".
- Created a package named "MMMMMMMMMMMMMMMMMMMMMM...".
- Had them own a file in a Differential revision, viewed that revision.
- Before: table is pushed out to several times the browser window width and everything is kind of a mess.
- After: package names get truncated to something reasonable.
{F5652953}
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13151
Differential Revision: https://secure.phabricator.com/D19478
Summary:
Ref T13151. See PHI684. Currently, the `MailableFunction` datasource does not include Owners packages, but they are valid subscribers and the `Mailable` datasource includes them.
Include them in the `MailableFunction` datasource, too.
Test Plan: Searched for revisions with particular package subscribers, got expected results in the UI (tokenizer knew about packages) and response.
Reviewers: amckinley, jmeador
Reviewed By: jmeador
Maniphest Tasks: T13151
Differential Revision: https://secure.phabricator.com/D19476
Summary: Ref T13151. See PHI616. There's a bug where the current banner changeset isn't cleared correctly when we hide the banner.
Test Plan:
- View revision with several changesets.
- Scroll down slowly through first changeset until banner appears.
- Scroll up until banner disappears.
- Scroll back down.
- Before: banner fails to reappear (code still thinks it's visible and we don't want to update it).
- After: banner reappears correctly.
Reviewers: amckinley, jmeador
Reviewed By: jmeador
Maniphest Tasks: T13151
Differential Revision: https://secure.phabricator.com/D19474
Summary:
Fixes T13145. The list controllers properly support public access already, but some of the view/detail controllers did not.
Allow logged-out users to browse builds, buildables, plans, etc., provided they can see the corresponding objects.
Test Plan: As a logged-out user, browsed around builds, build plans, logs, etc., without hitting any login pages.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13145
Differential Revision: https://secure.phabricator.com/D19459
Summary:
See PHI689. It can be difficult to distinguish between cards with the same number but different expiration dates (common when the bank sends you a new card).
For now, show the expiration date on the cart checkout screen.
Test Plan: Viewed a cart checkout screen with multiple cards, saw expiration dates.
Reviewers: amckinley
Differential Revision: https://secure.phabricator.com/D19462
Summary:
Fixes T13148. Ref T13105. The new document rendering engine for images let them overflow the UI bounds.
Add `max-width: 100%;` to keep them contained.
Test Plan:
- Viewed a very wide image in Safari, Firefox and Chrome. Saw sensible rendering.
- Also viewed a normal image, saw normal behavior.
Reviewers: amckinley, avivey
Reviewed By: avivey
Maniphest Tasks: T13148, T13105
Differential Revision: https://secure.phabricator.com/D19457
Summary:
Fixes T13147. In D19437, I changed this logic to support deleting the `""` (empty string) token, but `[].pop()` returns `undefined`, not `null`, if the list is empty and I didn't think to try deleting an empty input.
Fix the logic so we don't end up in a loop if the input is empty.
Test Plan:
- In any browser, deleted all tokens in a tokenizer; then pressed delete again.
- Before: tab hangs in an infinte loop.
- After: smooth sailing.
Reviewers: amckinley, avivey
Reviewed By: avivey
Maniphest Tasks: T13147
Differential Revision: https://secure.phabricator.com/D19456
Summary:
Depends on D19443. Creates a workflow for populating the new identity table by iterating over commits, either one repo at a time or all at once. Locally caches identities to avoid fetching them `inf` times. An actual migration that invokes this workflow will come in another revision that won't land until at least next week.
Performance is ~2k commits in 4.9s on my local machine.
Test Plan: Ran locally a few times with a few different states of the `repository_identity` table.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: jcox, Korvin, PHID-OPKG-gm6ozazyms6q6i22gyam
Differential Revision: https://secure.phabricator.com/D19446
Summary: Depends on D19429. Depends on D19423. Ref T12164. This creates new columns `authorIdentityPHID` and `committerIdentityPHID` on commit objects and starts populating them. Also adds the ability to explicitly set an Identity's assignee to "unassigned()" to null out an incorrect auto-assign. Adds more search functionality to identities. Also creates a daemon task for handling users adding new email address and attempts to associate unclaimed identities.
Test Plan: Imported some repos, watched new columns get populated. Added a new email address for a previous commit, saw daemon job run and assign the identity to the new user. Searched for identities in various and sundry ways.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin, PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T12164
Differential Revision: https://secure.phabricator.com/D19443
Summary: Depends on D19423. Ref T12164. Adds controllers capable of listing and editing `PhabricatorRepositoryIdentity` objects. Starts creating those objects when commits are parsed.
Test Plan: Reparsed some revisions, observed objects getting created in the database. Altered some `Identity` objects using the controllers and observed effects in the database. No attempts made to validate behavior under "challenging" author/committer strings.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin, PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T12164
Differential Revision: https://secure.phabricator.com/D19429
Summary: Ref T12164. Start building initial objects for managing `RepositoryIdentity` objects. This won't land until much more of the infrastructure is in place.
Test Plan: Ran `bin/storage upgrade` and observed expected table.
Reviewers: epriestley
Reviewed By: epriestley
Subscribers: Korvin, PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T12164
Differential Revision: https://secure.phabricator.com/D19423
Summary:
Ref T13142. When commits are pushed, we try to handle them on one of two pathways:
- Normal changes: we load these into memory and potentially apply Herald content rules to them.
- "Enormous" changes: we don't load these into memory and skip content rules for them.
The goal is to degrade gracefully when users push huge changes: they should work, just not support all the features.
However, some changes can slip through the cracks right now:
- If you push a lot of commits at once, we'll try to cache all of the changes smaller than 1GB in memory. This can require an arbitrarily large amount of RAM.
- We calculate sizes by just looking at the `strlen()` of the diff, but a changeset takes more RAM in PHP than the raw diff does. So even if a diff is "only" 500MB, it can take much more memory than that. On systems with relatively little memory available, this may result in OOM while processing changes that are close to the "enormous" limit.
This change makes two improvements:
- Instead of caching everything, cache only 64MB of things.
- For most pushes, this is the same, since they have less than 64MB of diffs.
- For pushes of single very large changes, this is a bit slower (more CPU) since we have to do some work twice.
- For pushes of many changes, this is slower (more CPU) since we have to do some work twice, but, critically, doesn't require unlimited memory.
- Instead of flagging changes as "enormous" at 1GB, flag them as "enormous" at 256MB.
- This reduces how much memory is required to process the largest "non-enormous" changes.
- This also gets us under Git's hard-coded 512MB "always binary" cutoff; see T13143.
- This is still completely gigantic and way larger than any normal change should be.
An additional improvement would be to try to reduce the amount of memory we need to use to hold a change in process memory. I think the other changes here alone will fix the immediate issue in PHI657, but it would be nice if the "largest non-enormous change" required only a couple gigs of RAM.
Test Plan:
- Used `ini_set('memory_limit', '1G')` to artificially limit memory to 1GB.
- Pushed a series of two commits which add two 550MB text files (Temporarily, I added a `--binary` flag to trick Git into showing real diffs for these, see T13143.)
- Got a memory limit error.
- Applied the "cache only 64MB of stuff" and "consider 256MB, not 1GB, to be enormous" changes.
- Pushed again, got properly rejected as enormous.
- Added `memory_get_usage()` calls to measure how actual memory size and reported "size" estimate compare. For these changes, saw a 639MB diff require 31,479MB of memory, i.e. a factor of about 50x. This is, uh, pretty not great.
- Allowed enormous changes, pushed again, push went through.
Reviewers: amckinley
Maniphest Tasks: T13142
Differential Revision: https://secure.phabricator.com/D19455
Summary:
Ref T13141. Currently, during first-time setup we don't surface all the details about connection exceptions that we could: the underlying exception is discarded inside cluster connection management.
This isn't a huge issue since the reason for connection problems is usually fairly obvious, but in at least one case (see T13141) we hit a less-than-obvious exception.
Instead, store the original exception and propagate the message up the stack so users have more information about the problem.
Test Plan:
- Configured an intentionally bad MySQL username.
- Restarted Apache and loaded Phabricator.
- Got a more helpful exception with a specific authentication error message.
{F5622361}
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13141
Differential Revision: https://secure.phabricator.com/D19454
Summary:
Ref T13137. See that task for discussion.
When we show a diff-of-diffs, we often render stubs for files which didn't change between the diffs. These stubs usually aren't a big deal, but for certain types of changes (like refactors) they can create a lot of clutter.
Instead, hide these stubs and show a notice that we hid them.
Test Plan:
- Created a revision affecting 4 files.
- Updated it with a diff that changed only 1 of the 4 files.
- Added an inline comment to a different file.
- Viewed the diff of diffs.
- Before: 4 changesets with two "nothing changed" stubs.
- After: 2 changesets with the stubs hidden.
{F5621083}
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13137
Differential Revision: https://secure.phabricator.com/D19453
Summary:
Ref T13137. The "analyze/cache data about changesets" step is becoming more involved. We recently added detection for generated code to support "Ignore generated changes" in Owners, and I now plan to hash the new file content so we can hide changes which have no effect.
Before adding this new hashing step, pull the "detect copied code" and "detect generated code" stuff out and move them to a separate `ChangesetEngine`. Then support doing a changeset rebuild directly with `bin/differential rebuild-changesets`.
This simplifies things a bit and makes testing easier since you don't need to keep creating new revisions to re-run copy/generated/hash logic.
Test Plan: Ran `bin/differential rebuild-changesets --revision Dxxx`, saw changesets rebuild. See also next change.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13137
Differential Revision: https://secure.phabricator.com/D19452
Summary:
Fixes T13140. See PHI660.
Recent versions of Subversion can send a `(get-file true false false )` protocol frame with extra space between "false" and "false". This is allowed by the protocol spec, but never normally happens, and we do not parse it correctly.
Instead, parse it correctly.
Test Plan:
- Added unit tests.
- Ran `svn proplist svn+ssh://.../diffusion/X/file.c` under SVN 1.10 before and after the change.
- Before: indefinite hang.
- After: completed in finite time.
Reviewers: amckinley, asherkin
Reviewed By: amckinley, asherkin
Maniphest Tasks: T13140
Differential Revision: https://secure.phabricator.com/D19451
Summary:
See D19446. This should make it easier to process larger, more complex result sets in constant memory.
Today, `LiskMigrationIterator` takes constant memory but can't apply `needX()` reqeusts or `withY(...)` constraints.
Using a raw `Query` can handle this stuff, but requires memory proportional to the size of the result set.
Offer the best of both worlds: constant memory and full access to the power of `Query` classes.
Test Plan:
Used this script to iterate over every commit, saw sensible behavior:
```name=list-commits.php
<?php
require_once 'scripts/init/init-script.php';
$viewer = PhabricatorUser::getOmnipotentUser();
$query = id(new DiffusionCommitQuery())
->setViewer($viewer);
$iterator = new PhabricatorQueryIterator($query);
foreach ($iterator as $commit) {
echo $commit->getID()."\n";
}
```
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19450
Summary: Ref PHI662. Viewing a dashboard you don't have permission to view (in the Dashboard application) currently fatals while building crumbs, since we fail to build the ` ... > Dashboard 123 > ...` crumb.
Test Plan:
- Viewed a dashboard I didn't have permission to view in the Dashboards application.
- Before patch, fatal when calling `getID()` on a non-object.
- After patch, sensible policy error page.
- Viewed a dashboard I can view, saw sensible crumbs.
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19449
Summary:
See <https://hackerone.com/reports/351361>. We currently require MFA on the screen leading into the user create flow, but not the actual create flow.
That is, `/people/create/` (which is just a "choose a type of account" page) requires MFA, but `/people/new/<type>/` does not, even though this is the actual creation page.
Requiring MFA to create users isn't especially critical: creating users isn't really a dangerous action. The major threat is probably just that an attacker can extend their access to an install by creating an account which they have credentials for.
It also isn't consistently enforced: you can invite users or approve users without an MFA check.
So there's an argument for just removing the check. However, I think the check is probably reasonable and that we'd likely prefer to add some more checks eventually (e.g., require MFA to approve or invite) since these actions are rare and could represent useful tools for an attacker even if they are not especially dangerous on their own. This is also the only way to create bot or mailing list accounts, so this check does //something// on its own, at least.
Test Plan:
- Visited `/people/new/standard/` as an admin with MFA configured.
- Before patch: no MFA prompt.
- After patch: MFA prompt.
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19448
Summary: Ref T13137. See PHI592. Depends on D19444. Apply a limit up front to stop patches which are way too big (e.g., 600MB of videos) from generating in the first place.
Test Plan:
- Configured inline patches in git format.
- Created a normal revision, got an inline git patch.
- Created a revision with a 10MB video file, got no inline patch.
- (Added a bunch of debugging stuff to make sure the internal pathway was working.)
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13137
Differential Revision: https://secure.phabricator.com/D19445
Summary:
See <https://discourse.phabricator-community.org/t/maniphest-home-page-crash-after-d19417/1445/3>. These special-token-only searches currently end up populating an empty `ownerPHIDs`, which fatals after the stricter check in D19417.
Make the fatal on `withConstraint(array())` explicit and only set the PHID constraint if we have some PHIDs left.
Test Plan: Searched for "No Owner", "Any Owner", an actual owner, "No Owner + actual user".
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19440
Summary:
See PHI652. When you `echo x | arc paste` today, you end up with a Paste object that has the empty string as its "language".
This is normally not valid. Pastes where the language should be autodetected should have the value `null`, not the empty string.
This behavior likely changed when `paste.create` got rewritten in terms of `paste.edit`. Adjust the implementation so it only adds the LANGUAGE transaction if there's an actual language.
Also, fix an issue where you can't use the "delete" key to delete tokens with the empty string as their value.
Test Plan:
- Created a paste with `echo x | arc paste`, got a paste in autodetect mode instead of with a bogus language value.
- Created a paste with `echo x | arc paste --lang rainbow`, got a rainbow paste.
- Deleted an empty string token with the keyboard.
- Deleted normal tokens with the keyboard.
- Edited subscribers/etc normally with the keyboard and mouse to make sure I didn't ruin anything.
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19437
Summary:
See PHI624. Some of the mobile navigation and breadcrumbs in support pacts aren't as good as they could be.
In particular, we generally collapse crumbs on mobile to just the first and last crumbs. The first crumb is the application; the last is the current page.
On `/PHIxxx` pages, the first crumb isn't very useful since the Support landing page is two levels up: you usually want to go back to the pact, not all the way back to the Support landing page.
We also don't need the space since the last crumb (`PHIxxx`) is always small.
Allow Support and other similar applications to tailor the crumb behavior more narrowly if they end up in situations like this.
Test Plan:
- With an additional change to instances (see next diff), viewed a support issue page (`/PHI123`) on mobile and desktop.
- Saw a link directly back to the pact on both mobile and desktop.
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19438
Summary:
Ref T13137. See PHI609. An install would like to filter audit requests on a particular branch, e.g. "master".
This is difficult in the general case because we can not apply this constraint efficiently under every conceivable data shape, but we can do a reasonable job in most practical cases.
See T13137#238822 for more detailed discussion on the approach here.
This is a bit rough, but should do the job for now.
Test Plan:
- Filtered commits by various branches, e.g. "master"; "lfs". Saw correct-seeming results.
- Stubbed out the "just list everything" path to hit the `diffusion.internal.ancestors` path, saw the same correct-seeming results.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13137
Differential Revision: https://secure.phabricator.com/D19431
Summary:
See <https://discourse.phabricator-community.org/t/maniphest-non-integer-point-values-in-csv-export/1443>.
We currently export the Maniphest "points" field as an integer, but allow it to accept decimal values (e.g. "6.25").
Also fix a bug where we wouldn't roll over from "..., X, Y, Z, AA, AB, ..." correctly for Excel column names if sheet had more than 26 columns.
Test Plan:
- Set a task point value to 6.25.
- Exported to text, JSON, XLS.
- Saw 6.25 represented accurately in exports.
- Exported an excel sheet with 27+ columns.
- Manually printed the first 200 column names to check that the algorithm looks correct.
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19434
Summary:
See PHI251. Ref T13137.
- Replace the perplexing text box with a checkbox that explains what it does.
- Mention this feature in the documentation.
Test Plan:
- Clicked/unclicked checkbox.
- Read documentation.
- Used an existing checkbox control in Slowvote to make sure I didn't break it.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13137
Differential Revision: https://secure.phabricator.com/D19433
Summary:
See <https://discourse.phabricator-community.org/t/hidden-description-field-in-maniphest-task-breaks-form/1432>.
If you hide the "Description" field in Maniphest, we still try to render a remarkup preview for it. This causes a JS error and a nonfunctional element on the page.
Instead, hide the preview panel if the field has been locked or hidden.
Test Plan:
- Hid the field, loaded the form, no more preview panel / JS error.
- Used a normal form with the field visible, saw a normal preview.
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19432
Summary:
See PHI638. When a diff is large (between 100 and 1000 files), we collapse content by default unless a change also has inline comments.
This rule isn't explicitly explained anywhere. Although it's not really a critical rule, it fits easily enough into the UI callout.
Also render the UI callout in a slightly more modern way and avoid `hsprintf()`.
Test Plan:
{F5596496}
- Also, clicked the "Expand" link and saw everything expand properly.
Reviewers: amckinley
Reviewed By: amckinley
Differential Revision: https://secure.phabricator.com/D19430
Summary:
Depends on D19427. Ref T13130. See PHI251. Support configuring owners packages so they ignore generated paths.
This is still a little rough. A couple limitations:
- It's hard to figure out how to use this control if you don't know what it's for, but we don't currently have a "CheckboxesEditField". I may add that soon.
- The attribute ignore list doesn't apply to Diffusion, only Differential, which isn't obvious. I'll either try to make it work in Diffusion or note this somewhere.
- No documentation yet (which could mitigate the other two issues a bit).
But the actual behavior seems to work fine.
Test Plan:
- Set a package to ignore paths with the "generated" attribute. Saw the package stop matching generated paths in Differential.
- Removed the attribute from the ignore list.
- Tried to set invalid attributes, got sensible errors.
- Queried a package with Conduit, got the ignored attribute list.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13130
Differential Revision: https://secure.phabricator.com/D19428
Summary:
Depends on D19426. Ref T13130. Ref T13065. While I'm making changes to Owners for "Ignore generated paths", clean up the "mailKey" column.
We recently (D19399) added code to automatically generate and manage mail keys so we don't need a ton of `mailKey` properties in the future. Migrate existing mail keys and blow away the explicit column on packages.
Test Plan: Ran migration, manually looked at the database and saw sensible data. Edited a package to send some mail, which looked good.
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Maniphest Tasks: T13130, T13065
Differential Revision: https://secure.phabricator.com/D19427
Summary:
Depends on D19425. Ref T13130. See PHI251. Now that changesets have a durable "generated" attribute, we can let owners packages check it when we're computing which packages are affected by a revision.
There's no way to actualy configure a package to have this behavior yet.
Test Plan:
- Created a revision affecting a generated file and a non-generated file.
- When I faked `mustMatchUngeneratedPaths()` to `return true;`, saw the non-generated file get no packages owning it.
- Normally: lots of packages owning it).
- Created a revision affecting only generated files.
- When I faked things, saw no Owners actions trigger.
- Normally: some packages added reviewers or subscribers.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13130
Differential Revision: https://secure.phabricator.com/D19426
Summary:
Ref T13130. See PHI251. Currently, changesets are marked as "generated" (i.e., the file contains generated code and does not normally need to be reviewed) at display time.
An install would like support for having Owners rules ignore generated files. Additionally, future changes anticipate making "generated" and some other similar behaviors more flexible and more general.
To support these, move toward a world where:
- Changesets have "attributes": today, generated. In the future, perhaps: third-party, highlight-as, encoding, enormous-text-file, etc.
- Attributes are either "trusted" (usually: the server assigned the attribute) or "untrusted" (usually: the client assigned the attribute). For attributes like "highlight-as", this isn't relevant, but I'd like to provide tools so that you can't make `arc` mark every file as "generated" and sneak past review rules in the future.
Here, the `differential.generated-paths` config can mark a file as "generated" with a trusted attribute. The `@generated`-in-content rule can mark a file as "generated" with an untrusted attribute.
Putting these attributes on changesets at creation time instead of display time will let Owners interact with changesets cheaply: it won't have to render an entire changeset just to figure out if it's generated or not.
Test Plan:
- Created a revision touching several files, some generated and some not.
- Saw the generated files get marked properly with attribute metadata in the database, and show/fold as "Generated" in the UI.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13130
Differential Revision: https://secure.phabricator.com/D19425
Summary:
Fixes T13135. See PHI633. For at least some video files with legitimate MIME type "video/quicktime", Chrome can play them but refuses to if the `<source />` tag has a `type="video/quicktime"` attribute.
To trick Chrome into giving these videos the old college try, omit the "type" attribute. Chrome then tries to play the video, seems to realize it can, and we're back on track.
Since the "type" attribute is theoretically only useful to help browsers select among multiple different alternatives and we're only presenting one alternative, this seems likely safe and reasonable. Omitting "type" also validates. It's hard to be certain that this won't cause any collateral damage, but intuitively it seems like it should be safe and I wasn't able to identify any problems.
Test Plan:
- Watched a "video/quicktime" MP4 cat video in Chrome/Safari/Firefox.
- See T13135 for discussion, context, and discussion of the behavior of some smaller reproduction cases.
Reviewers: amckinley, asherkin
Reviewed By: amckinley
Maniphest Tasks: T13135
Differential Revision: https://secure.phabricator.com/D19424
Summary:
Fixes T13132. I removed this branch in D19156 when tightening the logic for the new CSP header, but there's a legitimate need for it: downloading files via `arc download`, or more generally being an API consumer of files.
This is not completely safe, but attacks I'm aware of (particularly, cookie fixation, where an attacker could potentially force a victim to become logged in to an account they control) are difficult and not very powerful. We already issue clear setup advice about the importance of configuring this option ("Phabricator is currently configured to serve user uploads directly from the same domain as other content. This is a security risk.") and I think there's significant value in letting API clients just GET file data without having to jump through a lot of weird hoops.
Test Plan:
- With `security.alternate-file-domain` off, tried to `arc download` a file.
- Before: downloaded an HTML dialog page.
- After: downloaded the file.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13132
Differential Revision: https://secure.phabricator.com/D19421
Summary:
Ref T13130. See <https://discourse.phabricator-community.org/t/unable-to-create-owners-package-with-same-path-in-multiple-repositories/1400/1>.
When you edit paths in Owners, we deduplicate similar paths, like `/x/y` and `/x/y/`. However, this logic currently only examines the paths, and incorrectly deduplicates the same path in different repositories.
Instead, consider the repository before deduplicating.
Test Plan:
- Edited an Owners package and added the path "/" in two different repositories.
- Before: only one surived the edit.
- After: both survived.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13130
Differential Revision: https://secure.phabricator.com/D19420
Summary: See PHI611 for details.
Test Plan:
Ran a Buildkite build, saw Buildkite confirm receipt of these parameters in the HTTP response:
{F5562054}
Reviewers: amckinley
Reviewed By: amckinley
Subscribers: PHID-OPKG-gm6ozazyms6q6i22gyam
Differential Revision: https://secure.phabricator.com/D19419
Summary:
Depends on D19416. Ref T13110. Ref T13130. See PHI598. When rendering a "Very Large" revision (affecting more than 1,000 files) we currently compute the package/changeset ownership map normally.
This is basically a big list of which packages own which of the files affected by the change. We use it to:
# Show which packages own each file in the table of contents.
# Show an "(Owns No Changed Paths)" hint in the reviewers list to help catch out-of-date packages that are no longer relevant.
However, this is expensive to build. We don't render the table of contents at all, so (1) is pointless. The value of (2) is very small on these types of changes, and certainly not worth spending many many seconds computing ownership.
Instead, just skip building out these relationships for very large changes.
Test Plan: Viewed a very large change with package owners; verified it no longer built package map data and rendered the package owners with no "(Owns No Changed Paths)" hints.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13130, T13110
Differential Revision: https://secure.phabricator.com/D19418
Summary:
Ref T13110. Ref T13130. When a revision is "large" (100 - 1000 files) we hide the actual textual changes by default. When it is "very large" (more than 1000 files) we hide all the changesets by default.
For "very large" diffs, we currently still show the "large" warning, which doesn't really make sense since there aren't any actual changesets.
When a diff is "very large", don't show the "large" warning.
Test Plan:
- Viewed a small diff (<100 files), saw no warnings.
- Viewed a large diff (100-1000 files), saw just the large warning.
- Viewed a very large diff (>1000 files).
- Before: both "large" and "very large" help warnings.
- After: just "very large" warnings.
Reviewers: amckinley
Reviewed By: amckinley
Maniphest Tasks: T13130, T13110
Differential Revision: https://secure.phabricator.com/D19416
