quentin.pradet.me
I like maintaining things: the Elasticsearch Python clients by day, urllib3 and trustme by night. I write about #Python, HTTP, async/await, open source and performance! he/him
122 posts
1,095 followers
185 following
Regular Contributor
Active Commenter
comment in response to
post
Not familiar enough with weak refs to tell! Thanks for the tip.
comment in response to
post
I maintain the Python Elasticsearch client, and github.com/elastic/elas... is more complicated to fix than I thought. I'd be happy to provide more details if that sounds useful to you. (Not a lot of lines are affected, but it does require complex reasoning, mainly to keep backwards compatibility.)
comment in response to
post
Ah, I thought the initial examples were about the standard library. Using the fully qualified name (`better_dedent.dedent`) would clarify even if it’s uglier.
comment in response to
post
Thanks! For the record it starts at 19:03
comment in response to
post
The thing is I am more interested in how it works at a larger scale, which is supposed to be the point? Like adding a feature to an open source project with 100k lines of code or more.
comment in response to
post
Nice! I am a bit confused at the README which says 1/ the standard library handles t-strings fine but not f-strings and 2/ better-dedent is about t-strings.
comment in response to
post
The overhaul is a mess but we do have a clear plan to fix things. It will take time but the end result will be better.
Note that developers internally still write docs and complained as much as external users!
comment in response to
post
I should be working on that one issue but am planning that next trip!
comment in response to
post
I can recommend anything written by Martha Wells or Christopher Paoloni in the last 10 years. They both write fantasy and science fiction.
comment in response to
post
Thank you for sharing that idea with me, by the way. I should have attributed it to you!
comment in response to
post
Just like object-relational impedance mismatch, OpenAPI-based SDKs work well enough until they don't, at which point extraordinary effort is needed to connect the dots.
comment in response to
post
Another (major) issue: OpenAPI relies on JSON Schema, a data validation language. It is hard to go from a collection of oneOf and anyOf to actual request and response types in a given programming language. It works well enough for simple cases, so many projects still try.
comment in response to
post
I tried Python at some point but too active for me. And am considering joining CubeHead as I am getting into Rubik’s Cube to bond with my 6yo.
comment in response to
post
I try to join as few as possible because I can't keep things unread. But I have known students joining hundreds of them.
Right now I am on urllib3 (which is where we speak between maintainers, there is not much public activity) and Kludex (uvicorn and starlette).
comment in response to
post
Probably Discord, but then you have to join the relevant servers, not too unlike IRC actually
comment in response to
post
Another (minor) issue: you can't document enum members individually but need to describe them all at once: swagger.io/docs/specifi...
To avoid this, OpenSearch uses oneOf instead in its OpenAPI specification, but it's less explicit.
comment in response to
post
Right, I looked at the OpenAPI issue tracker, and indeed, nobody said that having optional resources was against REST principles.
They are still working on it, and it apparently raises all kinds of questions: github.com/OAI/sig-moon...
comment in response to
post
For example, OpenAPI does not support optional parameters in URLs! OpenAPI was initially designed for RESTful API, so mandating parameters made sense. Thankfully, version 4.x "Moonwalk" will fix that issue, but no known release date exists.
comment in response to
post
Mais effectivement tout semble intéressant !
comment in response to
post
I really needed this advice for a long blog post I am working on. Thank you!
comment in response to
post
Also not surprised. The amount of stuff you don’t have access to without an US location is large. This includes the Amazon iOS app, iPhone mirroring on macOS and the Waymo app.
comment in response to
post
Yes.
comment in response to
post
Probably? en.m.wikipedia.org/wiki/Linguis...
comment in response to
post
Après "mon chien a mangé mes devoirs" on a "ChatGPT a mangé mes devoirs" !
comment in response to
post
Tried this before with Soulwit ear pads but they only lasted a few months. :(
comment in response to
post
Cool idea! One data point about testing: as far as we can tell urllib3 works fine (as a pure Python package) but its tests rely on cryptography to generate TLS certificates. So we can’t test free-threading builds in CI.
comment in response to
post
Elastic always had offices around the world but they are mostly used for sales in practice and the whole engineering organization works remotely
comment in response to
post
I have to say seeing "momentum" and "anti-momentum" so close on my timeline was fun
comment in response to
post
(It also gives me AI-generated vibes)
comment in response to
post
I have stared at yours a lot (it's in every single blog post these days, for good reason) so I am biased but the professional one is… bland?
comment in response to
post
This is how Elastic launched Search, Observability and Security "labs", allowing engineers to publish whatever they want. Example: www.elastic.co/search-labs
This really helped as the process to write on the official blog is just too heavy.
comment in response to
post
Ah that’s more like it, thanks.
comment in response to
post
Probably just github.com/pypa/setupto...?
comment in response to
post
Sounds like a great opportunity, Sebastián 👍
comment in response to
post
Also it’s widely distributed, there’s an office in Amsterdam but most of the leadership is US-based
comment in response to
post
How would that lure users? LLM scrapers don’t need that to generate unsustainable traffic: lwn.net/Articles/100...
comment in response to
post
The culture is changing with our Serverless offering but even that is only released once per week: www.elastic.co/guide/en/ela...
comment in response to
post
Most of the Elastic Stack is released at the same time, with minor releases every ~5 weeks, the pressure to not miss feature freeze is real.
Some teams, including mine, try to get away from that to have more flexibility and release bug fixes immediately.
comment in response to
post
Probably not? But you can quickly switch between keyboards with a single tap on the earth icon. If you group langages wisely that could work well.
comment in response to
post
I did that initially but it sends the whole prompt again which is costly if there’s no context caching. It also anthropomorphizes LLMs needlessly.
It’s much better to evaluate the response with a thumbs up or down!