Pinned Tweet
But in a time when 50 megabytes of memory chips can be obtained for under $4000, there is less need for complex GC algorithms. Such techniques as reference-counting, ephemeral GC, closure analysis, etc., may not really be necessary now that it is possible to use massive memories.
A64m_qb0 retweeted
I added Idris 2 benchmarks: github.com/AndrasKovacs/smal… It came to light that both Agda and Idris 2 had quadratic parsing, only using non-operator functions and plain definitions. The Idris parser was fixed right away in this PR: github.com/idris-lang/Idris2…
A64m_qb0 retweeted
There are several different ways pattern-matches can go wrong in Haskell, and so GHC offers several different warnings about faulty pattern-matches. @RaeHaskell walks us through: youtu.be/Ferh1uLXOYQ
A64m_qb0 retweeted
Replying to @aspiwack
@Iceland_jack's github.com/ghc-proposals/ghc… is another design direction - it would allow you to write `fold xs @_ @(via Sum)`
Though for `fold` in particular at least one can write `ala Sum foldMap xs`
A64m_qb0 retweeted
But there are some thoughts in this direction in this paper (though I don't think that their proposed interface is quite right still) arxiv.org/abs/1807.11267v1 14/14
а, ну и сейчас уже куча людей пишут экстеншоны, которые включены по умолчанию, и которые включены другими экстеншонами, так что может и на треть не сократится даже
ну не знаю, насколько GHC2021 тут помогает
куча полезного туда не вошло (дерайвинг, рекорды, перегрузки, лямбда-кейс и т.д.)
ну на треть список сократится, но такое никто и не заметит если не сказать
A64m_qb0 retweeted
The expletive newtype deco-expletive-rations everywhere.
This is the real meat here.
Here is how you compute the sum of a list:
getSum $ foldMap Sum xs
or
coerce $ fold @_ @(Sum _) xs 8/14
A64m_qb0 retweeted
But sometimes you get to define a bunch of instances which are just a little bit too complex or specialised to be derived, but still mind-numbing for a human. And, I don't know, instance declaration take a lot of visual space and it ends up being a lot of clutter. 6/14
A64m_qb0 retweeted
The silly type-class-instance block.
This one doesn't happen everywhere. But most every time you define a type you have to equip it with a bunch of type-class instances. Most of them are boring. Some can be derived, when most are such, then you are good. 5/14
A64m_qb0 retweeted
A way forward for this would be the Local Module proposal, but it is far from completion github.com/ghc-proposals/ghc… 4/14
A64m_qb0 retweeted
The many lines of import statements.
They exist for a variety of reasons. Some good, some bad. They are annoying, pollute your git diffs a lot. But they are all at the top of the file, so it's not the worst. 3/14
ну не знаю, насколько GHC2021 тут помогает
куча полезного туда не вошло (дерайвинг, рекорды, перегрузки, лямбда-кейс и т.д.)
ну на треть список сократится, но такое никто и не заметит если не сказать
Replying to @aspiwack
The Long Litany of LANGUAGE Laments. Uh… I meant Pragmas.
This one is local, easily ignored, and largely alleviated by GHC2021. 2/14
A64m_qb0 retweeted
My Licentiate thesis just dropped!
The seminar itself is at 15:30 CET on Jan 17th, Zoom link available upon request (to deter spammers). After that, I'll be officially halfway done! In the meantime, enjoy ☺️
mpg.is/papers/gissurarson202…
A64m_qb0 retweeted
Я считаю, что пора прекратить делать идиотские переводы и заимствовать слова парсер, преттипринтер и тайпчекер. И пруфчекер заодно. Хотя бы для использования в профессиональной литературе.
A64m_qb0 retweeted
My next goal is to upstream External STG to GHC.
It will make the External STG Interpreter super easy to use.
A64m_qb0 retweeted
"[We find] situations where low-pause GCs ... achieve worse application latency [and much worse throughput] than simple stop-the-world GCs" arxiv.org/pdf/2112.07880
A64m_qb0 retweeted
I completely overhauled my older work on fast elaboration: github.com/AndrasKovacs/smal… It's faster, better, more documented. Here's a fun benchmark: smalltt loads an 1 million LOC file in 8.7s, coqtop loads the same thing in 439s.




















