P-Y

L8 Android Eng @Square | LeakCanary, Curtains, Radiography, square/logcat, square/PAPA

Public Key

npub19wwxc29uufy6rmegf9aza3hkcqdu00s8yyfyhr88mglptysgp3tsz7g5sa

NIP-05 Address

py_at_androiddev.social@mostr.pub

Profile Code

nprofile1qqszh8rv9z7wyjdpau5yj73wcmmvqx78hcrjzyjt3nna50s4jgyqc4cpz4mhxue69uhhyetvv9ujumt0wd68ytnsw43qxwfxvx

Publishing to

relay.mostr.pub

Author Public Key

npub19wwxc29uufy6rmegf9aza3hkcqdu00s8yyfyhr88mglptysgp3tsz7g5sa

Show more details

Last Notes

2024-09-20 23:25:38

- reply

Boarding, jfk to sfo. I might figure this out and build a very cool thing . Most likely I'll fall asleep though. Wish me luck

2024-09-20 23:24:34

- reply

I'm playing with rendering a heap dump as a treemap based on the heap's dominator tree. The issue is, most nodes end up having root as a dominator, making this not a very useful visualization. So I'm trying to figure out what's leading to most nodes being dominated by the forest root.

2024-09-20 23:20:54

Nothing like debugging heap graph traversal code at the airport with a cocktail & steak
#AndroidDev #dcnyc2024
https://cdn.masto.host/androiddevsocial/media_attachments/files/113/172/430/131/578/299/original/0958ebc9d2694e91.jpg

2024-09-17 04:50:28

SIXT: "Save time at vehicle pickup, Check-In now. Get free cancellation!"
Me: "sure why not, love free cancellation"
SIXT: "cool cool so all we need is send your FACEPRINT and ID to a 3rd party. It's cool they only keep it 3 years"
🤬
https://cdn.masto.host/androiddevsocial/media_attachments/files/113/151/079/372/156/650/original/4c78e1830ae36c33.png

2024-09-11 13:21:44

- reply

@npub1h2j…puef from reading the sources of Handler, Message, MessageQueue, View, ViewGroup and ViewRootImpl

2024-09-03 18:22:11

How do yall go about figuring out which broadcasts are system broadcasts (=> not exported), vs broadcasts from high privileged apps but not system, e.g. battery change (exported)?
Asking because Google Play Store is requiring targeting API 34 and.this dumb new rule of having to specify exported vs not exported for broadcasts seems tricky if you don't know which broadcasts should be exported.
#AndroidDev

2024-09-02 19:42:32

- reply

@npub1rks…hfs0
https://cdn.masto.host/androiddevsocial/media_attachments/files/113/069/658/311/536/420/original/dd7e27094b27527b.jpg

2024-09-02 15:12:25

Un canard a dit à sa cane:
« Ricane, ricane. »
Un canard a dit à sa cane:
« Ricane » et la cane a ri (LeakCanary).

2024-08-30 22:01:27

- reply

@npub14zr…9fx0 🫂

2024-08-17 22:21:30

I knew the Dagger coffee example would come in handy some day!
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/979/687/916/575/424/original/ae669b0e4eeaa5ff.jpg

2024-08-11 16:30:23

Does anyone here have the Skylight Calendar device? Any input from actual users?
I like the idea of a central cal on display in the living room. Ideally would show TODOs too or easy access.
I worry about the product being bought and / or shut down early (hi FB Portal 🖕).
Not interested in DIY (Raspberry PI, large android tablet, ..) unless it's low tech, I'm not looking for more side projects 😅
#Android #AndroidDev

2024-08-11 16:17:33

- reply

@npub159v…350f and yet, I did get some lovely answers from absolute strangers. Do you think randomly scolding people can achieve anything meaningful?

2024-08-11 13:54:26

We had this invasive plant that kept creeping in from a space in between our fence and the neighbors fence that I kept cutting back every few weeks.
We went away for 7 weeks and the thing went on a mission!
But also.. it grew Blackberries?! (left hand side of picture). I love blackberries! So I'm wondering if letting it grow is what brought the blackberries. Leave it be or cut it off?
#gardening
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/943/705/071/094/935/original/8dd47a64317f5470.jpg

2024-08-10 08:51:07

- reply

@npub174e…axnq et une fois que la médiane c'est intégré: "et la distribution, elle a quelle guelle?! Tu vas quand même pas comparer des médianes??

2024-07-24 09:40:34

SSO login has become so slow I probably wouldn't mind if it came with elevator music or a short video.
Actually I'm sure somewhere someone must be building a product to deliver information or ads while you wait on SSO
I'm looking at you, Okta. More like NotOkAtAll.

2024-07-12 13:49:35

UUuugh JIRA has a syntax for exact text search (insert escaped quotes, e.g. \"search this\") BUT but but.. it always performs stemming on search words anyway. E.g. if you do an exact search for "closing", you'll get all tickets that contain "closed".
The index only stores stems.
https://support.atlassian.com/jira-software-cloud/docs/search-for-issues-using-the-text-field/#Word-stemming
FUCK YOU JIRA.

2024-06-27 12:20:42

- reply

@npub187s…3zn5 loving the OG LeakCanary sticker 🥰🐤

2024-06-23 10:12:47

I went fishing. The only thing I caught? My phone, after it fell at the bottom of the pond.
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/665/390/559/640/646/original/ab53f92a344747d2.jpg
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/665/390/896/654/142/original/b974853e18e597f7.jpg
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/665/391/034/088/608/original/ba250b02f59ad9d1.png

2024-06-20 14:52:41

- reply

I just filed a ticket because I should probably follow up after yelling on the internets: https://issuetracker.google.com/issues/348333887

2024-06-20 11:35:14

J'ai souri, cool reference
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/648/735/459/486/505/original/f1eccfbef350bf98.jpg

2024-06-16 22:57:46

- reply

Cue the school WhatsApps group with 12 consecutive "voice chat" rows until someone finally warns others not to click on it.
Dear WhatsApp, this feature behavior is dumb as hell.

2024-06-16 22:56:03

Whatsapp has a "voice chat" feature, similar to Slack's huddle.
Problem: when you start a voice chat then stop, it leaves a "message" row that says "voice chat", with the duration the voice chat was active.
So folks in the WhatsApp group see this later, and, being unfamiliar with it, click on it thinking it's an audio message. Which restarts a voice chat with just them. They're alone so they leave after a few seconds. This leaves another message row that just says "voice chat". Rinse & repeat

2024-06-14 15:51:46

- reply

@npub14uk…2erc the ones that get me is when it's been a while and it feels like a thousand spider generations built a mega web and then I'm destroying all that work in one swipe

2024-06-11 15:49:07

What, no more gold 3D scrolling text?! Saving money on every possible corner?!
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/598/771/652/008/144/original/31c7c67c76880c03.png

2024-06-09 06:18:10

Start tracking?
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/585/203/810/084/918/original/b5e43d72546554c7.jpg

2024-06-07 03:33:49

AI WIFI: local gen AI that generates responses to your http requests without ever actually reaching the Internet.

2024-06-04 20:00:29

A quick marbré, which is really just a quatre quart (pound cake) where the top batter is mixed with 1/4 chocolate. Tastes best when the middle of thr chocolate is just set, barely cooked.
#MardiPatisserie
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/560/108/047/975/220/original/20f7d5222a11f4ee.jpg

2024-06-02 07:03:51

I finally reproduced a Recomposer test leak that has been making our UI tests flake for 2 years and forcing us to disable leak detection on a large number of tests 🎉🎉
https://github.com/android/compose-samples/pull/1409
Ok bed time now.

2024-05-31 06:26:09

Why yes, Datadog mobile can absolutely help you reduce your network traffic footprint.
See, it can track and log all network calls. But that won't be the most effective way to help reduce traffic.
Because, well, datadog mobile will upload close to 1 GB per device per day of monitoring data.
And that means, the best way Datadog can help you reduce network traffic footprint is by turning it off.
🙃

2024-05-30 23:23:09

- reply

New alpha with the changes in: https://square.github.io/leakcanary/changelog/#version-30-alpha-7-2024-05-30

2024-05-30 21:19:19

- reply

I mentioned that HeapDumpDeletionStrategy felt wrong since I don't always delete, and you really could do anything you want those heap dump files.
So, I'm renaming this to HeapDumpStorageStrategy
https://github.com/square/leakcanary/pull/2684
(this is all unreleased so I can rename things every other PR it's fiiine)

2024-05-30 21:11:20

- reply

Here's the fix: https://github.com/square/leakcanary/pull/2683/commits/d84896b8bbc3f9279d076acef7cea4213f76f107
Release coming later today!

2024-05-30 20:56:48

- reply

I think I'm ok paying that cost because looking up a class by name is rare in the first place, usually just an entry point.
Most of the time we're traversing a heap graph through references, so everything is optimized for that + low memory footprint.
"Low memory footprint" here means key design choices in critical places:
- Sorted arrays with binary search over hash maps and hash set (load factor=>25% wasted space)
- Byte arrays over objects

2024-05-30 20:14:28

- reply

That test is for later though: we don't even get to the assertion check here, as opening a heap dump preloads java.lang.Object, failing the test early.
We saw earlier that the issue lies with HprofInMemoryIndex, so now I can add another similar but more targeted test against HprofInMemoryIndex which checks the class name to class id reverse lookup.
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/531/867/206/019/283/original/a2091d4e832ab46e.png

2024-05-30 18:22:56

Things are great in this cave.

2024-05-30 16:32:12

- reply

So first, I reproduce this in a test, as my IDE & test environment makes debugging a lot easier.
Turns out, I don't even need to try doing anything, just opening a heap dump fails.
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/530/993/270/833/463/original/2775a5d48517b30f.png

2024-05-30 16:16:58

- reply

Of course, against my better judgment I skipped writing a test for this, and of course I messed up the logic. So I stopped, did not try to fix the bug, wrote failing tests, then fixed my bug.
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/530/934/643/956/157/original/5702c8f39cbb73c3.png
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/530/935/204/866/061/original/a8f15303f5e9a96a.png

2024-05-30 15:57:58

- reply

It's worth noting that usually LeakCanary runs on Android heap dumps. This is a JVM unit test, so it's a JVM heap dump, and these can be funky.
The first step to investigating is always reproducing. It doesn't repro locally, only in CI... and rarely. So I need that heap dump file that was created in CI.

2024-05-25 20:37:03

- reply

I should mention, when an existing system lacks automated tests, then its design is unlikely to make it amenable to lower level tests and it makes sense to start building the safety net at higher levels of integration. You just have to remember that it's just a start that provides you with some cover to start refactoring it into better & more testable pieces.

2024-05-25 16:55:54

- reply

So, again:
We write automated tests to help us catch bugs early and increase our confidence in our changes, both of which helps us move faster.
The value of a test, i.e. whether it helps us move faster, derives from balancing out its primary dimension of "finding bugs" against the other dimensions (fast, deterministic, unchanging, clear).

2024-05-25 16:54:05

- reply

Third: will it be unchanging, i.e. is it not brittle?
A brittle test is a test that fails or needs to be updated even though the external behavior of the code under test hasn't changed.
Strive to write unchanging tests: tests that only need to be updated if the external behavior changed but not after introducing refactorings, new features or bug fixes.
Fourth: is it clear?
Can I easily understand what the test is checking?
Can I easily diagnose failures?

2024-05-25 16:51:58

- reply

First, is it fast?
Faster test feedback means catching issues earlier or feeling confident about changes earlier, either way, we're moving faster
Second, is it deterministic, i.e. not flaky?
A flaky test is a test that fails nondeterministically, without any code change.
Flaky tests have a double tax: you need to rebuild AND you lose confidence in your tests so you slow down your rate of change to lower risks.

2024-05-25 16:49:51

- reply

At the core, an automated test exists because we believe it can catch bugs.
So, "will it find bugs?" is the primary lens of existence for a test. If the answer is "probably not", we can stop right there.
Now, assume we think yes, we need to remember that the actual question is "will this help us move faster". So we have to weigh "will it find bugs" against other factors, which will help us write better tests.
Let's look at those factors:

2024-05-25 16:46:15

- reply

Now, if we select developer velocity as our primary goal for testing, we avoid a lot of the classic pitfalls of testing (e.g. high reliance on integration tests or overspecified tests).
It's a lens that helps us remember that we don't test for the sake of testing, and it's a lens that helps us decide what makes for better tests.
And that's incredibly useful to help settle debates about this type of test vs that type, etc.

2024-05-25 16:41:51

- reply

Second, a bit less obvious:
Automated tests give us confidence that our changes are safe. This confidence allows us to make more changes and move faster. Move fast, don't break things!
Notable benefits:
- Enabling large refactorings
- Faster PR reviews
- Frequent releases

2024-05-25 16:41:03

- reply

Two reasons tests help with reducing the cost of making changes:
First, the obvious one: automated tests catch bugs before they impact our customers. The earlier a bug is caught, the cheaper it is to fix.
Just think about how much time it takes you to fix a bug when junit tells you right there that you messed up, vs when a customer calls with a vague description "it doesn't work".

2024-05-25 16:39:11

Why do we write automated tests?
This might seem obvious, but it's worth digging. Comments, questions and contradictions welcome!
Here goes:
*We write tests to reduce the cost of making changes.*
That means, to move faster. That's right, writing tests is first and foremost about increasing developer velocity.

2024-05-24 15:54:33

Shouldn't call them "fries" if they weren't fried at least twice. Looking at you In N Out 👀

2024-05-21 21:46:35

"Disagree & Commit"
...
git commit -am "Deleting this thing that shouldn't have been done" && git push --force

2024-05-17 03:34:54

- reply

If I could iterate on dominated nodes going from leaf dominated and progressively up, I could have each node add its own shallow+retained size just to its direct dominator, and things would automatically bubble up without having to recurse up chain.
As it turns out, I keep track of the list of nodes visited, in bfs order. So if I go through that list in reverse order and do a direct dominator size update for each node, that should work.

2024-05-17 02:37:46

- reply

So you end up with a map that has the same number of entries and same keys, but values are a very small set of objects. Which makes for much smaller up chains of dominators.
But I also did this in place: as I would go up to compress a chain, I'd compute retained size for all dominators in the chain and then remove those as processed, eventually skipping most nodes.

2024-05-16 18:36:17

The new profiler pane in Android Studio Koala looks really cool!
https://youtu.be/2wOfYgIMf-A?si=DaqFoUIvl1kINuIe&t=842
Also... LeakCanary might be coming to Android Studio in the near future 😘
#AndroidDev
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/452/203/341/979/967/original/3b03276ca672d604.png

2024-05-16 17:04:08

- reply

As I'm poking around tests, I look at the different layers of tests I have for ObjectGrowthDetector.
I have a test class that operates on in memory fake heap dumps, defined using a DSL and operated on as in memory byte arrays. It runs 12 tests in 100 ms.
I have a test class that operates on the heap of the test itself, leveraging hotspot heap dump APIs, which dumps into a temp file on the file system. It runs 10 tests in 9.5 seconds.
10 seconds vs 100 ms.

2024-05-16 16:23:27

- reply

And.. that's when I realize this is actually only used as an optimization to determine whether to run dominators and compute retained size. See, if we know we're going to do N analysis, we only need retained size on the last 2.
And that's when I spot a systematic off by 1 error in the uncommon case, and a start offset off by 1 error in the common case.
Tempted to fix it right there, but of course, no no no, gotta focus on writing the failing tests first.

2024-05-16 16:16:23

This morning, focusing on LeakCanary for a bit:
I think I've noticed a bug in how retained size increase for growing objects is computed in LeakCanary, because I noticed it seems to be 0 sometimes, included in cases where it's a very simple fake heap dump where it shouldn't be 0.
So I look at the tests for retained size.
Oh. No such tests 😬. Shame on me! While I do have tests for the dominator computation code, I don't have integration tests for how that's used in the heap growth code.

2024-05-16 06:23:15

- reply

One thing that's a bit counter intuitive is that `traverseHeapDiffingShortestPaths()` is a giant method, and so far I've resisted extracting methods to break it up. Why? Because doing so breaks my ability to move things around quickly, see patterns, handle edge cases globally, notice potential optimizations, etc. Which is exactly what I'm doing in this PR.. the cleanups will come when the time is right.

2024-05-16 06:08:59

- reply

Apparently 2000 characters is slightly over the max line length.
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/449/269/321/082/693/original/43778a17eb3d05d9.png

2024-05-16 06:02:30

- reply

@npub14zr…9fx0 I am as well, can't wait!

2024-05-16 05:49:41

- reply

How's your evening going? I writing heap dumps as hex string constants.
#AndroidDev
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/449/195/960/559/567/original/235428feda38dd17.png

2024-05-16 05:48:54

0b501e7e ca55e77e

2024-05-15 14:04:08

First day of Google IO was great starting from 3:30pm, i.e. when I stopped listening to any content and started talking with my Android friends.
Seriously, the general keynote AND the dev keynote were such a waste of time! It's not just the BS on AI, it's that it's incredibly boring, not tech oriented one bit., and repeating the same shit over and over. We get it, 2 Million tokens is more than 1 Million tokens. Jam those gemini gems in gemma.
Ugh.
#AndroidDev #GoogleIO

2024-05-14 16:37:20

WTF is happening right now
#googleio #androiddev
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/440/419/073/673/454/original/a53dfa389b76ce8a.jpg

2024-05-12 22:37:55

When you walk in circles to put baby to sleep and your Watch thinks you're starting a hike..
https://cdn.masto.host/androiddevsocial/media_attachments/files/112/430/512/361/038/798/original/f1890a1a6753978b.jpg

2024-05-09 04:11:28

Sometimes I think debugging software is hard..
but then I see our hardware engineers go into a cafe to figure out why our device touchscreen isn't working great there, they pull out a spectrum analyzer and sweep the outlets, and eventually find that spikes on the AM radio band are likely causing the issues... a local radio contaminating the earth ground of the AC power lines. WUT.
We have it easy folks.

2024-05-07 01:46:06

- reply

@npub1w3h…fsr2 if you're interested in making your code neither performant nor maintainable, and decisively weird, follow ME. https://github.com/pyricau/FunkyJFunctional

2024-05-07 00:53:15

New blog: Cutting some Slack, for
leaks and giggles
https://blog.p-y.wtf/cutting-some-slack-for-leaks-and-giggles
#AndroidDev

2024-01-14 19:06:19

- reply

@npub1upk…tcwy Related: folks often think that you have to choose between building tools for experts that provide a ton of capability & information but are incredibly hard to use & intimidating, or building tools for less experienced folks that are simplified by removing information & capability, rendering them effectively useless.
I think it's a false choice, and that, while significantly harder, we build tools that are useful & usable to both newcomers & experts.