my book is good and you should buy it

If you want this performance benefit, perhaps contribute in the GitHub PR saying you'd prefer the PR behavior. If cabal devs change their mind and unblock the PR, I'll get it ready and everyone can enjoy faster builds.

1) We're switching away from cabal to buck2 soon. So improvements to cabal aren't going to be relevant to us for much longer. 2) I don't really want to do work that isn't going to be accepted upstream and enjoyed by the whole community.

This change saves a few seconds on every cabal invocation in our (admittedly very large, 18k module now) cabal package. We have been using it with great success, but had to drop the patch because cabal had some other bug we needed to fix. I am considering redoing the patch, but-

The PR https://t.co/jiAQX7FNbz was blocked with implementing an "experimental features" facility into cabal, which does not exist, and the only reference or activity was from the meeting that blocked the improvement PR (linked above).

Want `cabal` to be faster? 🧵 Currently, the "preprocess modules" step tries all known preprocessors before looking for a `.hs` file. I have a PR that fixes this and looks for a .hs file first, and then preproccesor files if one isn't found.

https://t.co/1dtFEukVjG

The Subtle Footgun of TVar (Map _ _) wherein I talk about livelock, concurrent data structures, and the semantics of what a reference really means (link in next post)

Wrote it up! https://t.co/1dtFEukVjG

ok: `Map k v`, `IORef (Map k v)`, `MVar (Map k v)` bad: `TVar (Map k v)`, `TVar (Map k (Tvar v)` if you need transactionality you need `stm-containers` full-stop

need to write a blog post "TVar of Map is Always Wrong"

We're building something amazing at @mercury , and now it's available to the public. I got my own personal banking account and it's light years better than anything else I've used

"pick up artist" manual but for career advice

the design pattern where you bury the "login" menu and make "SIGN UP" incredibly prominent is terrible

Incrementalizing and decentralizing until I’m scattered on the winds

Lesson 8 has me learning important phrases like “I don’t know where the bathroom is” and “I want a beer”

doing Pimsleur lessons for Spanish while waiting for our tests to compile

brother.... we have 16k modules. only 1,358 extracted from main package. drawing boundaries with packages is very difficult to do without trashing the cross-package ux

"just don't have 12k modules" lol brother

GHCi scales pretty well up to about 12,000 modules, but after that, the linker and GC pauses are really punishing

love setting out to write a library, only to find that someone else did an optimal implementation a dozen years ago that is precisely what i need https://t.co/1287gcBlWe