Player is loading...
Metals: rich code editing for Scala in VS Code, Vim, Emacs and beyond
Ólafur Páll Geirsson, Scala Center
Wednesday, June 12, 2019 · 11:15 a.m. · 47m 15s
Rich editor tooling is a golden asset for any programming language, enabling beginners and experts alike to effectively build new applications and navigate existing codebases. Creating bespoke editor tooling is a challenge that that touches on a range of topics including compilers, build tools and user interfaces. In this talk we present Metals, a Scala language server that supports code completions, type hints, goto definition, fuzzy symbol search and other advanced code editing and navigation capabilities. Metals can be used in VS Code, Vim, Emacs, Atom and Sublime Text as well as any other Language Server Protocol compatible text editor. Metals works with the build tools sbt, Gradle, Maven and Mill thanks to Bloop, a fast Scala build server.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:01
oh hello everybody
00:00:08
so uh my name's olaf and i'm really excited to be here and present models
00:00:14
a project that provides rich code editing for skyline
00:00:16
text editors such is this call them emacs and beyond
00:00:21
um so briefly about myself i'm an engineer discuss enter our offices on campus here
00:00:28
uh which makes this moment even more special not welcome everybody here uh i'm the author of a code
00:00:34
from our calls colour twenty uh re factoring to cause cortex and metals the project that i'm presenting today
00:00:40
so you may have seen me as soon as a over p. g. young camp and traitor and
00:00:47
so or uh before i get started and does anyone know here but the skeleton or does
00:00:54
quite many hands which is great is better than a few years ago
00:00:57
uh so we're not for profit were established carry p. file on campus
00:01:02
and uh our two primary focuses on on open source and education uh so what i've been doing a
00:01:08
lot up the the signer is working on tooling uh and also if you may see on your badge
00:01:15
we're recognising salad days so um uh and also the the work that
00:01:21
understanding here is really not just me it's been a big collaborative effort
00:01:25
effort with a lot of contributors so uh one highlight hot here i was
00:01:29
working on the loop uh gabrielle into sitting here came from below no um
00:01:34
and mike into martial arts uh advertise lab and we're working together a full time since april
00:01:40
so uh and also wanna mention a few other names
00:01:42
you timber marco alexei you wanted some chain uh so um
00:01:50
before i wanna go and talk about text editing i wanna talk about another topic that we talk a lot about this coloured it's
00:01:56
a up discover centre which is how we grow this colour community uh and i think everyone here are flown all over the world
00:02:03
come um um the president's colour days should also be thinking about this question
00:02:08
is you know we have a scroll community in now in two thousand in nineteen
00:02:12
and and let's say we wanna double it in two thousand twenty one very ambitious goal how do we get there
00:02:19
um and it's really a difficult problem and and and there's no
00:02:24
silver bullet you're gonna have to have a lot of different solutions
00:02:28
so my thesis for the talk today is that improve that
00:02:31
italy is is one component um and i'll tell you why
00:02:36
i think the first question when you on board someone on your team they've never written scallop
00:02:42
the first thing you think about it is you know what editor do you use
00:02:46
what i'd be doing install um and if you are and what id to users colour
00:02:52
you fine for um some credit stack overflow core of people asking
00:02:56
the same question and it's been going on since two thousand and five
00:03:00
and and usually there he plea you know commented people like to have an opinion
00:03:06
uh i i i like this one here uh he says i'm a newbie whisk our
00:03:10
i'm developing a play framework i i've tried to call id based on equips an anthology
00:03:16
uh in my opinion the first one looks not mature yet and intelligent seems better
00:03:21
uh but i find it a bit slow i'm looking for the sublime option but it's not so documented so
00:03:27
i really like this comment because it highlights some of the points
00:03:30
i wanted to make its someone starting um with scarlet but has
00:03:36
like a particular goal i wanted to play up and and uh this person has tried to use different
00:03:42
ideas but is really looking for the sublime option 'cause i'm guessing that's what the person is familiar with
00:03:48
uh but if you've been long enough in in this call community you know that
00:03:53
the most uh put it answer is definitely intelligently without any explanation and
00:03:59
and that's also the yeah so um the two thousand eighties colour developer survey
00:04:05
and i kind of confirms that um eighty one percent of the respondents said that they use intelligent
00:04:10
um and this may not be surprising yeah for the people here in
00:04:13
the audience can i do a small surveyor who uses entirely different colour
00:04:19
yeah that's about me is not maybe first or so
00:04:26
i wanna take a moment here look at that the eighty one percent number because
00:04:33
if you take the global programming community it's not the same
00:04:38
so if you take the get a box covers the kind of their thirty one million programmers
00:04:43
globally and well then and that's a fair estimation i guess good abusers
00:04:48
number of of developers um and um the japanese developer survey
00:04:52
from two thousand eighteen says that they have six million people
00:04:55
who who use japanese products including intelligently and that's a really impressive number and and japanese products are
00:05:01
really fantastic uh but that still leaves twenty five
00:05:05
million programmers out there never even downloaded are installed intelligently
00:05:10
and um yeah so if you the the two thousand eighteen stack overflow developer survey
00:05:17
uh you also see a similar pattern which is that intelligent goes has twenty five percent
00:05:23
of the most popular that like a twenty five percent of the respondents say that today is there you know favour development environment
00:05:30
and it's up to six place there are five other editors
00:05:34
including notepad uh that uh you know are more popular um
00:05:40
yeah but also notable editor searches them sublime text and visual studio code so
00:05:47
there is an editor gap there like the representation of uh people use editors
00:05:50
in this colour community does not reflect how the global problem community uses editors
00:05:57
and i think that should be kind of a big concern going back to
00:06:01
the question how do we grow the community uh and and it's partly because
00:06:07
learning a new language is really hard unit learned syntax return semantics
00:06:11
i'm a bit learning a new language and menu editor with
00:06:15
all of the short cutting key bindings and everything is even harder
00:06:19
um and and this pretty much reflects how i started it's colour i started reading selling two thousand fourteen and i use the
00:06:26
flips from for quite a few months and and and i i chose that because i was familiar with the clips from java
00:06:32
any tips work right i learn scour uh was able to actually do a play at like the person read it
00:06:39
um and then eventually i switched intelligent uh because well it provided a
00:06:43
more feature rich experience um but what's really interesting is i switched intelligently
00:06:48
i didn't have to vance colour i was already familiar with that out of the box experience of using something that's
00:06:54
familiar so you can learn scholar faster i think is is one of the key components if you wanna grow the community
00:07:01
a student so few years go on um in two thousand i think i've actually started using
00:07:06
full time now obvious code uh with metals and and that's what i'm gonna be presenting today
00:07:12
it's now we can go back and talk about completions and her son and all
00:07:15
of the stuff that that this talk is really about it so i'm a quick tour
00:07:24
uh metals works with current leave officially supports five editors which means that
00:07:29
that's what we document our website works with the us call them emacs
00:07:35
adam and the button text so it works with any other language server compatible editor
00:07:40
but these are the ones that we have tested validated even at maybe some customisations
00:07:44
to really improve the user experience in these matters um and if you're using since last week or actually since the
00:07:51
official release last night from too much and we autumn off officially support all of these four build tools as well
00:07:58
meaning that if you just open up models and up workspace that has any of these build tools
00:08:03
it'll do all the installation automatically for you you don't even have to go online to the website
00:08:08
it just click of a button importers build any works so
00:08:13
and then once you've installed it it has a ton of functionality so all uh i'm not gonna go through all of this list here
00:08:20
so i'll just do a very brief them all just to give you a taste
00:08:26
so um what i have here is the the uh
00:08:31
uh metals and his coat open up in the aqua repository or and
00:08:39
some of the first things you'll see is there's um outline view here
00:08:42
you can walk through an and browse the classes you can fold some cold
00:08:49
and
00:08:52
you have completions and then we have some other fancy features like exhausted matches
00:08:59
and um
00:09:02
so that's pretty cool um and then i may show you uh we also show if
00:09:08
you hover over you get really nice information about the types um and i i think
00:09:14
this one is really helpful we show two parts one is the expression type which is
00:09:17
the type of that expression and and it's highlighted but it's a bit difficult to see now
00:09:22
we'll show you the generic signature of the method that you're
00:09:25
calling 'cause i think both both pieces of information really helpful
00:09:30
and and so if you if you hover over method that has some talk strings will also
00:09:34
give you a nicely formatted a doctor example or thank you
00:09:39
um and that we're suppose for german first column a and yeah you can go definition
00:09:45
you can find references and there i just clicked on and i found five hundred
00:09:48
thirty references to new symbol it was not at all cash uh we try to
00:09:53
the very fast with front references and so uh i just briefly another kind of out
00:09:59
of the box experience how it works i have another installation obvious code here that is um
00:10:06
has no extensions at all installed open up the just disqualify linen they'll build and
00:10:12
what happens is that it pops up and says and i can help you with uh this colour file extensions
00:10:18
and there are quite a few we don't have somebody don't yeah so we're not at the top yep um but if you install it um then them
00:10:27
everything should follow from there
00:10:29
fig open up discovered file
00:10:32
and it takes a moment for the server to start and uh oh yeah i had
00:10:38
already maybe in part of the builder so everything should just worked off of yeah right
00:10:43
um but you would have prompted me to impart to build a which is a male build
00:10:48
so let me see here so let's see that yeah right then
00:10:54
i give you a taste of how metals looks like
00:10:59
or
00:11:04
so as you saw there's a lot of features over fifteen twenty different features and i'm
00:11:09
not gonna go enumerate every one of them and and and show you how they work instead
00:11:14
i think it's more interesting to focusing go more in depth on two topics
00:11:19
so the first one that we struggle with and that was the first feature that you that we have to implement was diagnostics
00:11:26
and they are the red and the the yellow warnings we goals bad it telling you that you haven't here um
00:11:33
and the challengers to get them fast and and and have them relevant like we
00:11:37
should be helping you um and then the second case that you be doing completions
00:11:42
so uh also giving them low latency um but also
00:11:47
support them and you know a large number of positions
00:11:51
let's start with the errors um as you can see here we're kind of
00:11:56
shuffling between a file that doesn't parse and file that has a type there
00:12:03
piper says number should be of type string but it's actually in it
00:12:07
and and we answered the the string interpreter we're missing a closing curly brace
00:12:14
and essentially what this case studies about is just to to how we
00:12:20
tackle this problem of trying to provide these errors so that they're helpful
00:12:25
uh but um but do it fast in a fast
00:12:29
way um so there are two approaches actually do diagnostics
00:12:34
and i i would say one of them is you can just publish them from whatever you compile you run could be compiling you
00:12:40
belittle and then you scrape the logs and you put it into the editor and that's what i would call publishers from the bill tool
00:12:47
i'm an an an an alternative approaches that you have a presentation compiler like a library p. i.
00:12:53
where you do that compilation yourselves in the added in the i.
00:12:56
d. e. and you you you publish them directly yourself and the
00:13:01
the presentation of fire approach is pretty much the industry standard uh
00:13:06
it's used right holidays quality type script java language servers and so
00:13:12
models uses the one from the bill tool uh so
00:13:17
i'm gonna explain why we do that and why i think it's actually has really nice properties um
00:13:23
so in a nutshell when you type compiling s. p. t. it's gonna run compiling hundred twenty eight source files and then
00:13:29
give you an air because you have some problem and what metals there's a just
00:13:33
take that there was evident your editor uh and this has a very nice properties
00:13:40
it's essentially correct by construction whatever you have in your build
00:13:44
that environment is perfectly reflected in your editor there's never the case where
00:13:50
something like a dependencies off sync or anything uh because it's the same thing so if you're using you know
00:13:57
um and i'd scripts which which channel a different built all well we should get the air is directly from that source
00:14:03
uh instead of trying to reproduce that environment if uh and another one which i think is really exciting as
00:14:10
well is that because we perform that batch compilation so
00:14:13
frequently um we can share with your tests when you're running
00:14:19
so you as you're editing you're performing the c. p. u. cycles that are needed for you to run tests
00:14:26
anyways so whenever you go from your endurance your terminal you don't spend any moment compiling 'cause it's already compiled
00:14:33
so i think these two key components make their published like nothing from the build really appealing and and you um
00:14:41
and the first thing that you would expect wise doesn't everyone do this it sounds
00:14:45
fantastic well you get slow feedback as you're editing it may not be as interactive
00:14:52
and another downside as well is that you need to save the file
00:14:56
because the contents that you're editing it to be visible to the bill to
00:15:03
so let's tackle one of those two you know both of the the
00:15:06
downsides slow feedback it requires file save i'll start with the slow feedback
00:15:12
so having looked at this a bit and and we were looking at how to speed things up
00:15:17
the first lesson that we learn was publishing airs is pretty fast you essentially run that pipeline and
00:15:23
you the errors come streaming from the bill to let me just put them straight into the editor
00:15:28
um or is as fast as the bill can be yeah but another the one of the challenges
00:15:34
that removing them knowing that that that that are here is on can be a bit slow all um
00:15:41
so we're gonna look specifically into that how do you clear it so imagine that
00:15:44
we have three source files a. b. and c. and and eighty hasn't here before
00:15:51
and you run the batch compiler and and it runs in faces so it'll run the parser
00:15:56
phase and all three files and then rinse the type or face which is usually the expensive one
00:16:00
then advanced twenty more faces and then eventually you will reach the j. v. m.
00:16:05
and and let's everything's exceeds perfectly we know hey there are no more problems in this file then we can share it
00:16:14
and and that's kind of the naive approach if you say okay let's clear the air is you
00:16:17
have to wade through these all of these faces to finish problem files and that's a bit difficult
00:16:24
so uh what you could do instead is to start group the errors by the faces
00:16:31
that they're published in and then i just publish it after the parser faces finished um
00:16:37
which gives you a bit faster feedback but you know um
00:16:44
user pass or fail but but again you you're still waiting for the type or face to finish all of the sources before you can clear
00:16:51
so essentially is that what you could do is you can group years by whatever facing unit the finish
00:16:57
uh they pair from and and then you clear once those are finished
00:17:01
and this approach is implemented in loop so this is not implemented metals
00:17:06
uh but it's one of the key components that really sped up that you fix that air and it gets removed
00:17:12
and and this was a huge improvement i think in the user experience um and
00:17:18
was really great so metals communicates with loops through the
00:17:23
bill server protocol uh intelligent also has a client implementation
00:17:28
um and you can plug in any other build all that you weren't um
00:17:34
which i think is really great because it's accessible so people are using baylor pounds you could plug in another bill too
00:17:39
but also keep in mind that some of that custom work in terms of providing faster feedback that will
00:17:45
need to be implementing those bill tools as well that maybe gives you feel of of how metals works um
00:17:52
and and one thing that we did two two it was like a very cheap when in terms of giving you
00:17:58
faster feedback was we only probably just look at that
00:18:02
we can use like the presentation compiler approach for syntax errors
00:18:07
so you can run a parser you don't need to know the class pass you don't need to know any of the dependencies are the bill pipeline
00:18:13
and if there's a missing you know closing parentheses around and close
00:18:16
string literal we can just publish those right away um and as you
00:18:21
can see there's a typewriter below which was published from the bill tool
00:18:24
the the trick that metal does models that says that is able to
00:18:28
work really well with an old snapshot of a file that has a diaper and then we get new syntax errors and then we have to compute
00:18:35
hell they align into the new file um and what together working you're able
00:18:40
to get pretty snappy feedback as you're typing while keeping very stable type errors
00:18:45
so uh i think this also helps a lot with the slow feedback so i mean i'm
00:18:50
not gonna claim it's a solved problem uh it's probably still slower than what you used to but
00:18:55
for me personally in the project i'm working are which are small
00:18:57
to medium size it's pretty good and it has was really nice
00:19:03
attributes of crackers by construction insured compilation run and test so let's look at the last
00:19:08
piece which is requires file save so if you just typed in this thing and your buffer
00:19:15
which is an invalid program it has the specs a string that gets in it it should be an error but there is no air in the other
00:19:23
because you have to press save and then you get an or if you're coming
00:19:27
from an editor like clips or from tele today and this maybe you know big counterintuitive
00:19:35
and and what you can do is that's effects that are and and the air is still there well it's there because you haven't saved yet
00:19:43
so i i've seen that these people like that that out of the box experience opening of metals and like
00:19:48
well it's not what i expected it to be um having worked with that now myself for a few months
00:19:57
i'm no longer sure if it's really a feature or bug because
00:20:01
when you press save it triggers that type checking which is expensive it controls your c.
00:20:07
p. u. um and having control over exactly when that happens because you're editing you kind of
00:20:16
you're obviously adding an invalid problem otherwise you wouldn't be adding in the program
00:20:21
so most of the time it's broken anyways and most of the
00:20:24
time you know what's broken so uh i find the work flow where
00:20:28
you can just type without being distracted by moving errors chi you finishing
00:20:33
something is saying hey this is wrong and or a like an identifier
00:20:37
uh is relief nice i think in metals that you can just type it at
00:20:41
yours pays maybe you take a sip of coffee and and nothing's going crazy now
00:20:46
and then when you press a when you wanna get here um and and so i just recommended give it a
00:20:51
try but i just take the liberty to flip the downside into an upside and say it's a feature i saw it
00:21:02
so um that's that's a story around diagnostics and and there's more to it
00:21:06
but i think these were the ones that were kind of interesting and worth sharing
00:21:10
uh and the second one is is a feature that i think everyone
00:21:13
care deeply about completions uh so i've been working on models i think for
00:21:19
seven months and i wasn't using it for my data developmental still using intelligent
00:21:24
um because we were missing completions but uh once we had
00:21:28
completions along with parameter heads which are the ones that show you
00:21:31
the parameters that you're filling in and also being able to show the types of these all three came together in one release
00:21:39
and i was able to switch and i haven't really even had the urge to go back for a
00:21:43
lot of reasons um mostly because they're just really good rather than the really fast like fast so um
00:21:52
i'll explain how how how we do that so we use a presentation compiler approach so as i mentioned
00:21:57
before we use the bill tool for the air is good for the completions we run around presentation parlour
00:22:04
we need to know all the class pass and we need to reproduce that compilation but we have the
00:22:08
liberty of not having to publish any errors we can
00:22:11
do whatever we like 'cause giving you completions like false
00:22:16
like not giving you completions 'cause there's an air is not the end of the world um so so that approach works really great here
00:22:23
and i remember one before i started working on models i heard everyone talk about the presentation compiler does
00:22:29
like where do i had where we depend on this library and i didn't look and they didn't even
00:22:33
know what a classes colorado use it and there were no documentation that was no website about it so
00:22:39
if if you're curious it's a class called skeletal
00:22:42
fantasy interactive that global which extends the sour patch compiler
00:22:47
uh you can depend on it by depending on this colour compiler uh artifact
00:22:52
and it's used by the rebel i'm a grapple this
00:22:56
collide yeas colour fiddle um a lot of other tools
00:23:02
and it implements essentially two kinds of conditions one of them is the scope compilations where you just
00:23:07
pull out an identifier that's already imported in scope and number completions where
00:23:11
you would do a dog and a list of the members of that type
00:23:15
um and as i said before i've just been really pleasantly surprised at that
00:23:19
it's that way and it provides just really reliable member completions in scope completions
00:23:25
but for that reach completion experience
00:23:30
that i was used to intel a. j. school completion remember conclusions were not enough well i need what what
00:23:36
i was missing was quite a few other competitions and that's what metal it's implements on top of the presentation binder
00:23:44
so i'm gonna go through quite a few of the pieces and that we do to kind of make that experience which are
00:23:50
uh and the first one highlights also that it's not that
00:23:53
the presentation compiler should really even have these features because these features
00:23:58
require an editor 'cause you performers factory things we can explicitly move the cursor there's all sorts of
00:24:04
things that wouldn't really would that would be really difficult to encode in the library a. p. i. of the presentation
00:24:10
combined are essentially just gives you a list of symbols and it's up to us to figure out what to do
00:24:15
the first one is when you press enter and you select an item
00:24:19
um you can say this is the refractory should perform on the document
00:24:24
and one of the things that you can do is you can have sniff it's so which is essentially say
00:24:29
it shouldn't start opening parentheses move the cursor inside of the parentheses when you finish that you should trigger
00:24:35
parameter it's that's what you see when you select a greater which is another method
00:24:41
it should have any parentheses and you select filter it moves the cursor into that
00:24:46
filter parentheses and when you select next it knows that there are no arguments it moves the cursor at the end of it
00:24:53
so it's pretty basic but it's still kind of an important come piece of that dated experience of writing code
00:25:02
so one of the fancy features that we have is exhausted match completions
00:25:07
if you do a match completion it'll provided to say just a regular
00:25:11
match or an exhaustive match what we do is this works for um
00:25:18
uh both seal types as well as not seal types which
00:25:22
is different if you're coming from ontology so if you have
00:25:26
a seal type still baby tee it works for both whatever you haven't scope but it'll also work even if those
00:25:32
and nana sound even if they're not in in the the scope
00:25:36
we would out important for you which is what you would expect but
00:25:40
if it's not a sealed eighty t. if it's just like an exception
00:25:43
we will uh not would be don't exhaustively much all of that um
00:25:48
sometimes of exceptions but um i i yeah i'm sorry i forgot to be we give you all
00:25:55
of the exceptions when you do the case completions so i'll just jump into the next one so um
00:26:01
one thing that we do here which is really nice as well is that not only exhausted matches
00:26:06
but individual case statements and i think this is really a feature that i never had an intelligent and
00:26:11
i i really injury models is that you're doing a collect and you're just taking one of the cases
00:26:16
and this is the one that works with both seal types and also for exceptions 'cause
00:26:21
it would just provide you the completions or one of the exceptions that you haven't scope um
00:26:26
um i find this to be really useful um and another polish is
00:26:31
just a small thing but it's one of these things that required a
00:26:33
bit of of thinking was that if you have tasted need and we
00:26:37
know that none matches the type of the qualifier about match so we'll
00:26:43
sorted out the top but we don't filter out the rest because it might be that you wanna do nail dot something else are we it's
00:26:49
really that would be inexpensive analysis the just the sorting is customised and
00:26:53
tweak them polished and so the next one which i was desperately missing which
00:27:00
uh was over reading methods implementing something from an interface
00:27:05
um and in metals you can do that one method at a time you go into
00:27:08
the body like the template of that class here extending seek which has like five hundred methods
00:27:15
yeah but only three of them are abstract so when you do the iterate or you can see that the interior
00:27:20
sort of the the top because that's the one that we should implement first uh but then we go back
00:27:25
and and then there it's all the the concrete once and we automatically know that should be over a keyword
00:27:30
um and we also compute the correct kind of it's a c. given
00:27:34
so then it should be a room filter within it to hold and um
00:27:41
the next one which is a very important features well and it's really saving a lot of keystrokes is
00:27:45
oh i'm gonna need to access that a. p. i've paths um and as you can of surveillance or a an import
00:27:52
at the top so if you if you come from and holiday you take it for granted
00:27:55
it just works perfectly um but this is not part of the presentation compiler so we um
00:28:03
i had to implement it ourselves and the way we insert
00:28:06
the imports is nowhere near as sophisticated as intelligent as it
00:28:10
we just take that symbol only imported at the bottom of the global
00:28:14
import list so you will have to manually go up there and sort them
00:28:19
according to become edges of your project a really nice does all that on the
00:28:22
road that we just it's it's already really good as it is a better than nothing
00:28:28
so the next step is that you do and i think this is the piece that i cared a
00:28:31
lot about because i hate on it all the time is that you have conflicting names you have files
00:28:35
already imported but you wanna use another a. p. i. which also has
00:28:38
the name files because library developers use the same names all over the place
00:28:44
uh and it's really frustrating if files would only show was the file that is in scope metals
00:28:51
always does a full search on the whole thing every completion is exactly the same when we wouldn't
00:28:57
filter out something and there aren't any smart completions that you have to do with another shortcut to
00:29:02
get these results it's just a single command complete we do the full search so if you're gonna do
00:29:09
calm go go while files we know well hey this is already a consequence symbol
00:29:13
in school will just fully qualified for you and i find this really nice um
00:29:19
so another conclusion that we do which i from the
00:29:23
kind of scope a number completion prospective makes no sense
00:29:28
is that you you wanna do a school completion on the code as if it's written inside of a string literal
00:29:34
um so this is kind of a cue thing it didn't take a lot of work to do to implement but you have to
00:29:40
i understand that you're inside of a string literal you have to do your own parsing to
00:29:43
figure that's of the identifier you pull out the scope of that code identifier in the new
00:29:48
you you figure that and then you have to perform the restructuring to insert the
00:29:53
the u. d. u. leading ass and also you have to skip to my dad
00:29:56
of characters in the string and and sometimes you have to put the curly braces
00:30:01
uh and then once you have the argus you can also select the member which was just done the same
00:30:06
way that i think this one my favourite features to implement it was a lot of fun um and very useful
00:30:14
so
00:30:16
you seen all the features now um and there are more but but i think these are the ones that are most interesting
00:30:23
so a lot of customisations um on the feature adding more
00:30:27
richness but we also added a few customisations to try to
00:30:30
squeeze out as much performance as we could because its latency sensitive as you're
00:30:35
typing you wanna get the feedback as fast as you as possible so these are
00:30:40
i would play in the biggest contributors to having decent performance if you're gonna do completion
00:30:46
with the form of the presentation power um i will go through each of them individually
00:30:51
so if you beers a benchmark doing
00:30:55
it's actually a six thousand line file uh doing completions
00:30:59
on on um i'm a member will do dot and um
00:31:05
no the first line is essentially we just did a new
00:31:08
presentation compiler would just gave it to full class pass from zero
00:31:13
and it does that completion uh with a hop j. v. m. so that does ninety milliseconds so if
00:31:19
you just keep around that compiler instance and then you do another completion an a in in a different file
00:31:27
um then it will be forty milliseconds so you
00:31:32
shave off fifty milliseconds just by reusing that compiler
00:31:35
so this seems like you know a no brainer when um but you also have to be careful that the longer you use it
00:31:44
the more symbols get added to the symbol table and you it's likely that you make it more more wrong results
00:31:49
so it's a kind of a delicate balance between trying to reset it um
00:31:54
and reusing and what happens usually is that you get requests in bursts like
00:31:58
five completions in roll all in the same source files to parameter hands three hovers
00:32:03
uh and then we reuse but then metals very aggressive that reciting it again and again and again um but
00:32:10
it really says the most of the requests were able to shave off those fifty millisecond overhead which is significant
00:32:17
uh another i'd say biggest contributor to being also robust and uh and performance is that
00:32:24
and so it's collagen have compiler plug into this p. t. it's by ad compiler plugin
00:32:29
and people have lifters workroom over all sorts of stuff um metals essentially
00:32:35
has a white list is to approved component reckons if you're using kind projector
00:32:42
we'll out if you're using better melodic far we're about to using any other plug in
00:32:48
sorry we disabled it
00:32:52
so if you have a so and it's from the observation
00:32:56
the most compatible is actually don't change the semantics of the language
00:33:01
so these two actually introduce new like if if
00:33:04
you wouldn't have kind projector the conclusions would break
00:33:08
if you don't have bred a better manic for they would also breaking where grace a bit less obvious
00:33:13
ways so we disable everything else including macro paradise so
00:33:18
micro paradise also qualifies for the changes the language semantics
00:33:22
but it also transforms the trees that we get so much that all of the positions are broken
00:33:28
so the completion don't work anyways if ah and then uh
00:33:34
you may be wondering now on using come back rotations my project does that mean
00:33:38
it's totally broken and that's not at all the case that's really cool observation is that
00:33:43
if you have my connotations in a project the conclusions were totally fine with them the whole we'd don't work if you have
00:33:50
an annotation in a source file and you're trying to complete in the same source file
00:33:55
for those numbers so if you have another project or any other another
00:33:59
separate file or their expand that they will appear in your conclusions it's only
00:34:04
if you're in the same file uh but that wasn't your trade off because
00:34:08
you missed a few completions in that case but if we enable that you'd miss
00:34:12
almost all of them anyways because the trees are broken anyways so that was kind of
00:34:18
how we ended up another one that we do is is we actually disable black box macros we don't expand them
00:34:25
and and this saves from some benchmarks in like a source i with two hundred
00:34:28
lines very small on doing some fast course parsers and i was able to shave off
00:34:33
around hundred and thirty milliseconds if i remember correctly by just turning them off uh and
00:34:38
if you're not familiar with box markers there is actually some code where you you right
00:34:43
let's say here you wanna generate the time stamp up the compilation time so what happens is that the compiler will
00:34:50
invoke this call it well executed at compile time and put in a new syntax tree based on what you expanded
00:34:58
and just observe here that the it returns along the tree but it's annotated doesn't any
00:35:05
um i'll explain that in just a bit but we like what macros have this really nice
00:35:12
via attribute that you can disable them and they don't affect completions in anyway
00:35:17
so the completions are totally accurate even if you discard them you just getting
00:35:22
a hundred fifty millisecond shave after your type checking which is awesome um there's
00:35:27
another flavour macros that are called white box and there's no way we like
00:35:30
if we disable then we would produce incorrect results and i'm just gonna briefly explain why so if you're right
00:35:36
if you have the original colour which would be something markers dot times up if it was a black box micro gets expanded into the tree
00:35:43
but because the method hasn't any type description the expansion and then we'll wrap ascribe that expression to any
00:35:50
so um you it doesn't leak that it was a long but
00:35:55
uh why possible not to do that type script type description meaning
00:35:59
that the x. will have the type long meaning than the type of x. will influence the completions the rest of the source file
00:36:06
so this means if you're using models with colour remake shapeless that has h. least generated by macros
00:36:12
uh you will get the correct h. list from like a person with the initiative string in it than a journal
00:36:18
um but just keep in mind that you're paying also for the cost here
00:36:24
um another big when is just by not doing something requests of all that's an even
00:36:30
better you confirmed to down to zero um so imagine that we have a request queue
00:36:37
uh probably been moving my hands in the wrong direction the whole time 'cause they're
00:36:41
the screen here is wrong but anyway so the request you goes this way and we're
00:36:47
responding to the the first completion here and in the meantime well there is money
00:36:52
to the other senses three new uh completions um once we're finished with the first completion
00:37:01
we cancel the two that are older and we start working on the latest completion
00:37:07
uh and this was only chains now in the last release that was coming last night in the sense that we previously we would
00:37:13
respond to them in order and what we could have this getting up like a relaunch you of requests and you wouldn't get feedback
00:37:20
now we just prioritise whatever you just submitted and that makes sense in
00:37:24
a world where you're editing you only care about the last completion that
00:37:27
you triggered not the one that was before and this is great because we shave off like we should be skip a bunch of requests um
00:37:36
and what we can do is well is the editor can
00:37:40
send us cancel this and so even in while is in progress
00:37:46
uh metals if the other tells us cancel this we will be able to turn of an even stop the type checker
00:37:51
during middle of type checking thanks to the cancellation supporting the compiler that that's really great
00:37:58
and the last piece is a feature that is just implemented in the
00:38:02
compiler in the presentation the binder or you can instead of type checking all
00:38:06
source file you can give it a position and it will just type check down to that position and you get the types tree at that position
00:38:14
so here just as an example we're doing we're completing numbers and that means we need to common type checked the object models
00:38:21
we get numbers but it's as a type description list of it so we don't have to type check the one in the
00:38:28
the body of that method and so this works out of the box if you just use the presentation opener
00:38:33
for scope and i'm a member completions but but all
00:38:37
the customisations extensions that we do with actually we need
00:38:40
to work a lot with the compiler trees and they're half height checked so if you're working on the code
00:38:46
base you come in you need to be aware that it's really easy to have a node types on some trees
00:38:51
and you need to be aware that yours should use just targeted type big and and and that's pretty
00:38:56
much all models does now and it means it makes a huge difference i don't have any numbers but
00:39:00
for the parameter hands uh it went from just like oh i i wanna i don't wanna do
00:39:05
prominence 'cause it was so slow to the point where it was just always there and i didn't bother
00:39:10
and so that makes it made a huge difference um so as i said before models as a lot of features and we had
00:39:19
to um to cover two case studies of diagnostics and completions there's a lot more to
00:39:26
it so feel free to come to me and and asking the questions if you're curious
00:39:31
and i'm gonna just very quickly go over the others that we support
00:39:36
but it'll be a bit fast since i i don't have so much time left um
00:39:42
if you're wondering what's up with the hyper on visual studio code i'll give it you know in in a minute so
00:39:49
it has really great support for other languages that are very widely used primarily type scripting playful like it's fantastic
00:39:56
sophie using those languages committees collect you probably wanna just continue using these code
00:40:01
it has i was piece of work out of the box which means that it's
00:40:04
just built into the editor and the view school team maintains that support which is great
00:40:09
um i'm gonna show i don't think i'll have time to show you but i wanted to show you a there are two
00:40:14
features called life share an remote development which are just fantastic
00:40:18
it allows you to per problem with the remote caught code collie
00:40:21
and you're just editing as if you're on the same computer in the cool drive style
00:40:25
collaboration uh and it's fantastic remote development allows you to s. s. h. into the remote machine
00:40:32
i know as in your obvious coding you just say on open up a folder in an s. s. h. remote
00:40:38
and then you just editing on that machine that you have the grade you why would
00:40:42
know fiddling with file watches or anything just works out of the box and it's fantastic um
00:40:49
it has a great extension ecosystem and having written worked a bit on the metals
00:40:53
extension that we've ridden type script as well and i just enjoy writing
00:40:57
the extension it takes me like you have good completions apart you press play
00:41:03
you get like a new fresh instance in two seconds that has that version of whatever you change so it's really snappy in great
00:41:09
and i've tried to do it with the the other editors is just the best i've seen interviews code
00:41:14
um and it's very actively developed like every month there new releases and
00:41:18
i'm i look forward to reading the release notes because it's just something exciting
00:41:23
so the only really bad thing i can say is that the
00:41:25
syntax selling first colour kind of socks uh it's going way better
00:41:30
thanks to a contributor compare now for the past weeks and i'm just so thankful to woodward because
00:41:38
ah previously was is especially batter oh string interpreters
00:41:41
um and non ascii identifiers and stuff like that
00:41:45
it's so i'll mention just as good this because the time is out um uh that life sure um
00:41:53
is really cool so give it a try and the role the remote development where
00:41:57
you can use as you know you can code on a machine with a hundred course
00:42:02
and two hundred gigs of ram and it just works seamlessly there's no it's like you
00:42:06
just do it from the us code you why it's fantastic to give it a try
00:42:10
um so for them it's legendary model editing i've used it
00:42:15
myself for python development for a while and i love that uh
00:42:20
it has i'm just gonna say it wasn't the case a year ago but to date has a comprehensive list the support in fact
00:42:27
or two different implementations like two different and plug ins the comprehensive
00:42:32
implement most of the spec which is just fantastic um and i'd say
00:42:40
the biggest complaint i can have having tried to use it now with metal says that
00:42:45
if you wanna say hey would you like to import the build that is a blocking u. i.
00:42:49
widget would be really nice to just have a thing that's on like non blocking on the side
00:42:54
and you can continue editing a click on the button whenever you want and
00:42:56
and this is really really annoying uh and if you haven't seen it there's an
00:43:03
in development i don't recall only been which kind of solves that problem but
00:43:07
still keeps all the goodies of him so i'm personally reciting about that one
00:43:12
uh emacs i don't know much about so don't shoot me
00:43:16
but it seems to be legendary customisable uh and it
00:43:21
also not the case a year ago but it has to be a comprehensive list the
00:43:25
support which is just fantastic and uh it supports pretty much all of the features that we
00:43:30
supporting models including the advance parts around completions um
00:43:35
and i would say it's the same issue as of him but probably
00:43:39
some of the signal that's not the case you just need another plugin um
00:43:43
and then if you wanna have like a non blocking you i've aged to do imports it's kind of invasive
00:43:49
and sublime text i've never used really but i have like great respect for
00:43:54
it's primarily because it's legendary lightweight and it has really good
00:43:59
the falls in my opinion um and it's not as comprehensive today is election them
00:44:06
um the l. s. p. package but it actively developed and i think there's an
00:44:09
open p. writing some the more advanced pieces in completions so it's getting there really close
00:44:14
um which makes a very optimistic i think is great i would really we should one text
00:44:19
like that's sensible defaults philosophy is you just have a list the out
00:44:23
of the box and would be a huge boost for for someone an
00:44:27
i just pick one thing i'm sure there are other things but we have commands like import building currently have to like copy paste jason
00:44:34
config every person has to do that there's no way for us
00:44:36
to or maybe there is but i haven't figured out and lastly adam
00:44:42
uh it seems to be legend there next as well um it has
00:44:47
really these nice you averages the same that b. s. code has but
00:44:52
sorry maybe someone is i mean uh the l. s. b. support does not
00:44:57
uh the the completions don't handle some the mormons re factoring so we do um
00:45:00
no i'm not aware of any current p. r. six in these issues so like hello
00:45:05
imports like the imports just don't appear which is really about out of the box experience
00:45:10
not recommend this if you're showing it to beginners and say just install metals news item
00:45:15
like some stuff doesn't work and there's nothing there's no way for us
00:45:18
to to to figure work around that um so that's it so future work
00:45:24
as you can imagine an i. d. is a huge task uh
00:45:29
running and testing and debugging we have my neck who is from versus levels been working on this
00:45:35
full time now for over a month i think and i'm really excited i'm i'm optimistic we'll be able
00:45:40
to have just run and maybe down the road even debugging uh but also run tests and you could see
00:45:46
the passed passed on the field tests and i i'm really excited about that i'm getting close to that
00:45:52
out of the box experience for beginner who was not doesn't even have to know what's colour justices or something
00:45:58
and doesn't have to figure out the yeah with the incantation for just testing the single unit test
00:46:03
it's all of the testing memories of different syntax and winning the symbol
00:46:07
we're pretty much there or was we have find references and a bit requires
00:46:10
a little more custom development to to to be correct i'm re factoring support
00:46:16
is nonexistent at the time uh i lust he has great infrastructure for restructurings
00:46:22
uh but we just don't implement at the moment and it's a big project
00:46:27
i work six i think uh would be fantastic to have if you just open up the file you can see the values in line
00:46:33
well they are as you just learning anyway p. i. and i would estimate this is not even hard to add uh with
00:46:39
our current infrastructure and the same thing is pretty much the
00:46:42
same thing for s. b. t. and and and scholar script files
00:46:46
uh so probably a hundred other features that could be added i'm better
00:46:51
at these are the highlighting ones i think are most exciting thank you
Conference Program
A Tour of Scala 3
Martin Odersky, Professor EPFL, Co-founder Lightbend
June 11, 2019 · 5:15 p.m.
8337 views
A story of unification: from Apache Spark to MLflow
Reynold Xin, Databricks
June 12, 2019 · 9:15 a.m.
1267 views
In Types We Trust
Bill Venners, Artima, Inc
June 12, 2019 · 10:15 a.m.
1569 views
Creating Native iOS and Android Apps in Scala without tears
Zahari Dichev, Bullet.io
June 12, 2019 · 10:16 a.m.
2232 views
Techniques for Teaching Scala
Noel Welsh, Inner Product and Underscore
June 12, 2019 · 10:17 a.m.
1296 views
Future-proofing Scala: the TASTY intermediate representation
Guillaume Martres, student at EPFL
June 12, 2019 · 10:18 a.m.
1157 views
Metals: rich code editing for Scala in VS Code, Vim, Emacs and beyond
Ólafur Páll Geirsson, Scala Center
June 12, 2019 · 11:15 a.m.
4695 views
Akka Streams to the Extreme
Heiko Seeberger, independent consultant
June 12, 2019 · 11:16 a.m.
1552 views
Scala First: Lessons from 3 student generations
Bjorn Regnell, Lund Univ., Sweden.
June 12, 2019 · 11:17 a.m.
577 views
Cellular Automata: How to become an artist with a few lines
Maciej Gorywoda, Wire, Berlin
June 12, 2019 · 11:18 a.m.
386 views
Why Netflix ❤'s Scala for Machine Learning
Jeremy Smith & Aish, Netflix
June 12, 2019 · 12:15 p.m.
5026 views
Massively Parallel Distributed Scala Compilation... And You!
Stu Hood, Twitter
June 12, 2019 · 12:16 p.m.
958 views
Polymorphism in Scala
Petra Bierleutgeb
June 12, 2019 · 12:17 p.m.
1113 views
sbt core concepts
Eugene Yokota, Scala Team at Lightbend
June 12, 2019 · 12:18 p.m.
1656 views
Double your performance: Scala's missing optimizing compiler
Li Haoyi, author Ammonite, Mill, FastParse, uPickle, and many more.
June 12, 2019 · 2:30 p.m.
837 views
Making Our Future Better
Viktor Klang, Lightbend
June 12, 2019 · 2:31 p.m.
1682 views
Testing in the postapocalyptic future
Daniel Westheide, INNOQ
June 12, 2019 · 2:32 p.m.
498 views
Context Buddy: the tool that knows your code better than you
Krzysztof Romanowski, sphere.it conference
June 12, 2019 · 2:33 p.m.
394 views
The Shape(less) of Type Class Derivation in Scala 3
Miles Sabin, Underscore Consulting
June 12, 2019 · 3:30 p.m.
2321 views
Refactor all the things!
Daniela Sfregola, organizer of the London Scala User Group meetup
June 12, 2019 · 3:31 p.m.
514 views
Integrating Developer Experiences - Build Server Protocol
Justin Kaeser, IntelliJ Scala
June 12, 2019 · 3:32 p.m.
551 views
Managing an Akka Cluster on Kubernetes
Markus Jura, MOIA
June 12, 2019 · 3:33 p.m.
735 views
Serverless Scala - Functions as SuperDuperMicroServices
Josh Suereth, Donna Malayeri & James Ward, Author of Scala In Depth; Google ; Google
June 12, 2019 · 4:45 p.m.
936 views
How are we going to migrate to Scala 3.0, aka Dotty?
Lukas Rytz, Lightbend
June 12, 2019 · 4:46 p.m.
709 views
Concurrent programming in 2019: Akka, Monix or ZIO?
Adam Warski, co-founders of SoftwareMill
June 12, 2019 · 4:47 p.m.
1974 views
ScalaJS and Typescript: an unlikely romance
Jeremy Hughes, Lightbend
June 12, 2019 · 4:48 p.m.
1377 views
Pure Functional Database Programming‚ without JDBC
Rob Norris
June 12, 2019 · 5:45 p.m.
6375 views
Why you need to be reviewing open source code
Gris Cuevas Zambrano & Holden Karau, Google Cloud;
June 12, 2019 · 5:46 p.m.
484 views
Develop seamless web services with Mu
Oli Makhasoeva, 47 Degrees
June 12, 2019 · 5:47 p.m.
785 views
Implementing the Scala 2.13 collections
Stefan Zeiger, Lightbend
June 12, 2019 · 5:48 p.m.
811 views
Introduction to day 2
June 13, 2019 · 9:10 a.m.
250 views
Sustaining open source digital infrastructure
Bogdan Vasilescu, Assistant Professor at Carnegie Mellon University's School of Computer Science, USA
June 13, 2019 · 9:16 a.m.
375 views
Building a Better Scala Community
Kelley Robinson, Developer Evangelist at Twilio
June 13, 2019 · 10:15 a.m.
245 views
Run Scala Faster with GraalVM on any Platform
Vojin Jovanovic, Oracle
June 13, 2019 · 10:16 a.m.
1342 views
ScalaClean - full program static analysis at scale
Rory Graves
June 13, 2019 · 10:17 a.m.
463 views
Flare & Lantern: Accelerators for Spark and Deep Learning
Tiark Rompf, Assistant Professor at Purdue University
June 13, 2019 · 10:18 a.m.
380 views
Metaprogramming in Dotty
Nicolas Stucki, Ph.D. student at LAMP
June 13, 2019 · 11:15 a.m.
1250 views
Fast, Simple Concurrency with Scala Native
Richard Whaling, data engineer based in Chicago
June 13, 2019 · 11:16 a.m.
624 views
Pick your number type with Spire
Denis Rosset, postdoctoral researcher at Perimeter Institute
June 13, 2019 · 11:17 a.m.
245 views
Scala.js and WebAssembly, a tale of the dangers of the sea
Sébastien Doeraene, Executive director of the Scala Center
June 13, 2019 · 11:18 a.m.
661 views
Performance tuning Twitter services with Graal and ML
Chris Thalinger, Twitter
June 13, 2019 · 12:15 p.m.
2003 views
Supporting the Scala Ecosystem: Stories from the Line
Justin Pihony, Lightbend
June 13, 2019 · 12:16 p.m.
163 views
Compiling to preserve our privacy
Manohar Jonnalagedda and Jakob Odersky, Inpher
June 13, 2019 · 12:17 p.m.
302 views
Building Scala with Bazel
Natan Silnitsky, wix.com
June 13, 2019 · 12:18 p.m.
565 views
Diversity and Inclusion Panel: Bring the Thunder!
Panel
June 13, 2019 · 2:30 p.m.
245 views
Asynchronous streams in direct style with and without macros
Philipp Haller, KTH Royal Institute of Technology in Stockholm
June 13, 2019 · 3:45 p.m.
304 views
Interactive Computing with Jupyter and Almond
Sören Brunk, USU Software AG
June 13, 2019 · 3:46 p.m.
681 views
Scala best practices I wish someone'd told me about
Nicolas Rinaudo, CTO of Besedo
June 13, 2019 · 3:47 p.m.
2707 views
High performance Privacy By Design using Matryoshka & Spark
Wiem Zine El Abidine and Olivier Girardot, Scala Backend Developer at MOIA / co-founder of Lateral Thoughts
June 13, 2019 · 3:48 p.m.
754 views
Immutable Sequential Maps – Keeping order while hashed
Odd Möller
June 13, 2019 · 4:45 p.m.
277 views
All the fancy things flexible dependency management can do
Alexandre Archambault, engineer at the Scala Center
June 13, 2019 · 4:46 p.m.
389 views
ScalaWebTest - integration testing made easy
Dani Rey, Unic AG
June 13, 2019 · 4:47 p.m.
468 views
Mellite: An Integrated Development Environment for Sound
Hanns Holger Rutz, Institute of Electronic Music and Acoustics (IEM), Graz
June 13, 2019 · 4:48 p.m.
213 views
Closing panel
Panel
June 13, 2019 · 5:54 p.m.
400 views
