The key observation to me was that the traditional Datalog/Prolog way of unifying is through syntactic equality, which is a bit too simple to express the kind of equality needed in Rust and elsewhere. You can express it in Datalog, but as it gets farther away from the source, error-generation suffers.
"Insightful observation" doesn't quite do it justice ;) There's a whole discipline in discrete math dealing with the possibilities and limitations of equational theories and reasoning. And it could be said their limitations gave rise to the various constraint formalisms that were introduced in the 80s and 90s as Prolog extensions and sometimes (syntactic) generalizations.
Well then, please elaborate because I know nothing of the discipline you describe and it certainly sounds interesting (though it also sounds a little fussy).
I'm not going to summarize classic equational reasoning (with it's deep connection to algebraic geometry and whatnot) here in a single HN post ;) I can point you to some classic works/authors in the field, though (and I'm sure some fellow HNers can provide some more): Gauss, Knuth, Bendix, Gröbner, Buchberger, Bachmaier, Euclid, Ganzinger, Davis, Putnam, Robinson, and Colmerauer for Prolog 2. There are also category theory papers relevant to reasoning about data structures, and of course Damas, Hindley, Miller for type theories. The way is the goal here.
>> I'm not going to summarize classic equational reasoning (with it's deep connection to algebraic geometry and whatnot) here in a single HN post ;)
Aw :(
Alright, I'll just chase down some of the references you say. I was going to check out Prolog 2 anyway, after browsing the wikipedia article on Colmerauer a few days ago and seeing a reference to his later work there.
To be a bit more concrete than the grandparent's name soup, I liked this handbook article about unification in theories: Franz Baader and Jörg Siekmann. Unification Theory. In D.M. Gabbay, C.J. Hogger, and J.A. Robinson, editors, Handbook of Logic in Artificial Intelligence and Logic Programming, pages 41-125. Oxford University Press, Oxford, UK, 1994.
>> The key observation to me was that the traditional Datalog/Prolog way of
unifying is through syntactic equality, which is a bit too simple to express
the kind of equality needed in Rust and elsewhere.
I was surprised by this comment because Horn logic, although syntactically
restricted, has Universal Turing Machine expressivity and it's difficult to
see what Rust needs to do that needs more expressive power than a UTM(!).
Reading the post I understand that the author is speaking about unification in
particular, not Horn clauses and resolution, where unification is typically
used, however there is still a subtle error, which is the identificatin of
unification as "syntactic equality". While unification can be used to decide
whether two terms are syntacticall equal, in the sense that if they are, they
will unify, unification on its own is not syntactic equality.
First of all, say that T₁ = f(χ) and T₂ = f(α) are two terms. Note that χ is a
variable and α is a constant. Clearly T₁ ≠ T₂ and that is true for "syntactic"
equality, because χ ≠ α.
Now suppose that θ = {χ/α} is a substitution of the variables in {T₁,T₂}.
Now, θT₁ = Τ₂. Read θT₁ as "T₁ with θ applied". In other words, T₁ is equal to
T₂ _if_ χ is substituted with α. A unification algorithm (normally, Robinson's
unification algorithm) is used to establish that two terms are equal given a
substitution of their variables. In particular, a substitution of variables
that allows two terms to unify is a "unifier". Note that unifiers can be
composed, so for example, if T₃ = f(χ,υ), T₄ = f(α,β), φ = {χ/α} and ψ = {υ/β}
then φT₃ ≠ T₄ but φψT₃ = T₄.
I noticed in the examples of Rust-in-Prolog given in the link above that the
author subtly fudges the syntax of Prolog, for example Bar, which is normally
a variable in Prolog, is used as a constant, whereas ?A and ?B which would be
not-variables in Prolog (but partially ground terms) are used as variables. I
wonder if this is part of a now-divide-by-zero step in the author's reasoning,
that I can't figure out exactly because I don't know any Rust. For example,
what is "?A" and what is "Bar" in the context of Rust?
In any case, the author is of course correct that unification is not type
equality- but then, nothing is except for type equality. As far as I
understand these things anyway- I'm not a type nerd. Bottom line, you can't
implement "type equality" as any other kind of "equality" and you'll have to
jump through hoops -the hoops of defining your higher-level notion of
equality- in what ever language you choose.
> you'll have to jump through hoops -the hoops of defining your higher-level notion of equality- in what ever language you choose.
Right. The author is just disappointed that Prolog does have a built-in "equality" that isn't the equality they want. A nice solution in Prolog would be to write a meta-interpreter or to do term rewriting on the input program. This would allow the author to write
Yes. I just don't see why this is such a big hassle. Particularly because -I'm guessing- whatever the Rust compiler does to decide type equality can't be less hassle-y than that.
But, I might misunderstand the problem since, again, I don't know Rust.
In the Rust ecosystem, I believe the next-generation IDE support is already beginning to incorporate some ideas like this, although I believe it uses Prolog rather than Datalog (I'd be interested to learn more about how they differ). The logic library is called Chalk ( https://github.com/rust-lang/chalk ), and it's designed to be a full trait resolution engine for Rust (trait resolution being a necessary step to e.g. autocomplete methods on types); rust-analyzer has been using it for quite a while now, and it's slowly being incorporated into rustc itself. Not as impressive as basing the entire IDE on queries, of course!
The Rust project is also using the differential-datalog library mentioned in the OP to underlie their third-generation borrow checker: https://github.com/rust-lang/polonius
> I believe it uses Prolog rather than Datalog (I'd be interested to learn more about how they differ).
Datalog is the positive, function-free fragment of Prolog, i.e., what Prolog offers over Datalog is negation and “value invention” through function terms. One consequence of that is that deciding whether a Prolog program entails a fact is undecidable (i.e., Prolog is Turing-complete), deciding whether a Datalog program entails a fact is can be done in polynomial time with respect to the size of the database.
Fantastic work! Datalog/Prolog are old-yet-futuristic technologies that I wished every developer knew about, as it solves a lot of problems in very elegant ways (as attested to by this post).
I noticed you (assuming you're the author) used what looks like a non-traditional Datalog syntax – which makes sense to me as, IMO, Datalog/Prolog desperately need first-class support for a record-like syntax to finally break into mainstream. Is there any prior work to this syntax, or did you just develop it as you needed it?
There's a proposal due to the prolog-commons initiative for a dictionary-like object (that is also implemented in SWI Prolog, I believe).
And in Prolog you can, of course, just trivially use your own term structure for item-values:
p([ item : "value", ... ])
You could even use JSON-like terms in Prolog (with maybe a little help by the op/3 directive to sort out parsing priorities):
A_Prolog_Term = {
x: y,
z: [a, b, {1, d} ]
}
But the more fundamental approach IMHO would be to use nested knowledge-base terms:
p(
q(whatever).
r(X) :- q(X).
).
As pioneered in the 1980s for attribute grammars, using ^^ and other special graph tokens as nested "implies" operator, though (see "definite-clause translation grammars").
Agreed on the niceness of the keyword syntax, I would very much like to have than in Prolog too. SWI-Prolog has record-like data structured called dicts (IIRC) but you can't use them directly as clause heads the way the article does.
Both suffered from too heavy hardware requirements for what most companies were willing to pay for, otherwise we could have had Smalltalk like tooling, with incremental compilation and reloading for C++ already during the early 90's.
This is very cool! It's really awesome to see this code as data concept gaining a lot of traction recently. Hope to see this project developed further.
I'm working on a similar project here: https://sourcescape.io/, but intended for use outside the IDE on larger collections of code (like the codebase of a large company.)
Agreed on the Prolog/Datalog approach of expressing a query as a collection of facts. CodeQL does the same. From one datastore nerd to another, I actually think this is a relatively unexplored area of querying knowledge graphs (code being a very complex, dense knowledge graph.)
Very excited to see where you go next with this "percolate search" functionality in the IDE.
It is a queryable database of semantic information about the program which is generated by the compiler (compiler plugin, to be precise). Once generated, other tools which need semantic information, like linters or language servers, can consume it without having to worry about how to actually generate it.
You might enjoy a talk about it: How We Built Tools That Scale to Millions of Lines of Code by Eugene Burmako
I would have expected a mention of Roslyn which powers Visual Studio (Not code).
Sort of similar to the IntelliJ approach but it is also what drives the C# compiler, it maintains the code model (actually 2 of them, syntactic and semantic) which makes it pretty powerful.
It is multi language (VB, C# and I think F# now too?) but it’s perhaps not as universal as the Language Server Approach
One challenge to this that languages which easily allow metaprogramming can encourage codebases where it is hard to write tools to gain insight into the structure of a codebase.
You can say “don’t do that”, but I didn’t. Past coworkers did.
Are there any examples of "programs as databases" being applied to structured text like JSON or YAML? Something like a generic system which takes a set of rules/facts and the source data, and transforms these into a queryable data structure.
I'm working with a lot of OpenAPI [1] specifications currently, some of which span tens of thousands of lines. Heaps of parent, child, sibling type relationships, and objects which are defined-once and referenced in many places. It would be nice if I could perform a search like "select all schemas used in this path" or "select the count of references to this parameter".
Well yes, but that doesn't solve the parsing problem right?
My point is that some YAML or JSON documents conform to a specification which can be codified as a set of rules, which combined with the data form facts. I'm asking if there are existing systems to parse these files given I write these rules in a particular syntax, spitting out a queryable representation of the data.
This is extremely cool, great work! You're right, IDEs can do so much magic but they're limited by the 'spells' their creators have imagined. Many times I've wanted to do a simple semantic refactor that I could express as logic but the IDE can't handle it.
I don’t see what that has to with the IDE. Just grab a parser/ast library for your language.
I’ve written a few of my own refactorings for big design changes that might touch e.g. every view in a huge application in ways no series of regexes could cover.
It’s hard work. Definitely one of those tricky “cost of automation” decisions. You can do an awful lot of dumb grunt work in the time it takes to debug a refactoring.
This is how Rust Analyzers intellisense works, it uses a query engine called Salsa that stores the symbols in a database and the linting and completion/semantics are entirely query-driven:
I've written an IDE in JS where all internal structures are available in the runtime which can be hooked into. I can however not see the usability in what the article describe. So I would like to se some practical examples. What if you could do queries like in the article, what sort of queries would you run !?
Nice article and great idea, but as is traditional there are some slight
fudges of what Datalog is or what Horn clauses are etc, that I'd like to
unfudge, slightly. It's Sunday! What better than to start our day with a very
quick and almost not completely fudgy intro to logic programming? Le'ts go!
To begin with, Datalog is not a "cousin" of Prolog as stated in the section
"Interlude: Brief Intro to Datalog". Datalogs (there are many variants!) are
subsets of Prolog. For example, a typical datalog is the language of definite
clauses with no function symbols [¹] and with no negation as failure [²].
Another datalog may allow only the cons function in order to handle lists;
etc.
Otherwise the syntax of datalog is identical to Prolog, but there is a further
difference, in that Prolog is evaluated from the "top down" whereas Datalog is
evaluated from the "bottom up". What that means is that given a "query" (we'll
come to the scare quotes in a moment) Prolog will try to find "rules" whose
heads unify with a literal in the query (A literal is an atom, or the negation
of an atom; "p(χ,α)" is an atom.) whereas datalog will first generate the set of
all ground atoms that are consequences of the program in the context of which
the query was made, then determine whether the atoms in the query are in the
set of consequences of the program [³]. The reason for the different execution
model is that the bottom-up evaluation is guaranteed to terminate [⁴] whereas
Prolog's top-down evaluation can "go infinite" [⁵]. There is of course
another, more subtle difference: Prolog can "go infinite" because of the
Halting problem, from which datalog does not suffer because, unlike Prolog, it
does not have Universal Turing Machine expressivity [⁶].
So in short, datalog is a restricted subset of Prolog that has the advantage
of being decidable, while Prolog in general is not, but is also incomplete
while Prolog is complete [⁷].
Now, the other slight fudge in the article is about "rules", "facts" and
"queries". Although this is established and well-heeled logic programming
terminology, it fudges the er fact that those three things are the same kind
of thing, namely, they are, all three of them, Horn clauses [⁸].
Specifically, Horn clauses are clauses with a single positive literal.
Crash course in FOL: an atom is a predicate symbol followed by a set of terms
in parentheses. Terms are variables, functions or constants (constants are
functions with 0 arity, i.e. 0 arguments). A literal is an atom, or the
negation of an atom. A clause is a disjunction of literals. A clause is Horn
when it has at most 1 positive literal. A Horn clause is a definite clause
when it has exactly 1 positive literal.
The following are Horn clauses:
¬P(χ) ∨ ¬P(υ)
P(χ) ∨ ¬Q(χ)
Q(α)
Q(β)
In logic programming tradition, we write clauses as implications (because ¬A ∨
B ≡ A → B) and with the direction of the implication arrow reversed to make it
easier to read long implications with multiple premises. So the three clauses
above are written as:
And those are a "query", (a), a "rule", (b) and two "facts", (c) and (d).
Note that (b,c,d) are definite clauses (they have exactly one positive
literal, i.e. their head literal, which is what we call the consequent in the
implication). Facts have only a positive literal; I like to read the dangling
implication symbol as "everything implies ...", but that's a bit
idiosyncratic. The bottom line is that definite clauses with no negative
literals can be though of as being always true, hence "facts". Queries, i.e.
Horn clauses with no positive literals, are the opposite: "nothing implies"
their body literals (my idiosyncratic reading) so they are "always false".
Queries are also called "goals". Finally, definite clauses with both positive
and negative literals can be thought of as "conditionally true".
Prolog and datalog programs are written as sets of definite clauses, i.e. sets
of "facts" and "rules". So, when we want to reason about the "facts" and
"rules" in the program, we make a "query". Then, the language interpreter,
which is a top-down resolution theorem prover [⁹] in the case of Prolog, or
bottom-up fixpoint calculation in the case of datalog [¹⁰], determines whether
our "query" is true. If the query includes any variables then the interpreter
also returns evey set of variable substitutions that make the query true.
In the example above, (a) has two variables, χ and υ and evaluating (a) in the
context of (b,c,d) would return a "true" result with the variable substitution
{χ/α,υ/β}, i.e. (a) is true iff χ = α and υ = β.
And that's how Horn clauses and definite clauses become "rules", "facts" and
"queries".
Next time: how the leopard got its stripes and the hippopotamus learned to
love the first order predicate calculus.
_________________
[¹] This is my (first) slight fudge because constants are also functions, with
0 arguments. So, to be formal, the typical datalog above has "no functions of
arity more than 0".
[²] Negation-as-failure makes a language non-monotonic, in the sense that
introducing new "facts" can change the meaning of a theory, i.e. a program.
[³] So, its Least Herbrand Model, or its Least Fix-Point (LFP).
[⁴] Because it finds the LFP of the query and the program.
[⁵] Unless evaluated by SLG resolution, a.k.a. tabling, similar to memoization.
[⁶] Although higher-order datalogs, that allow for predicate symbols as
terms of literals have UTM expressivity, indeed a UTM can be defined in a
higher-order datalog fragment where clauses have up to two body literals with
at most two arguments:
[⁷] Less fudgy, definite programs are refutation complete under SLD
resolution, meaning that any atom that is entailed by a definite program can
be derived by SLD resolution. A definite program is a set of definite clauses,
explanation of which is coming right up.
[⁸] Long time ago, I explained this to a colleague who remarked that all the
nice syntactic elegance in query languages falls apart the moment you try to
make a query, which usually has a very different syntax than the actual rows
of the tables in the database. So I said "that's the point! Queries are also
Horn clauses!" and his immediate remark was "That just blew my mind". It's
been so long and I'm so used to the idea that I haven't a clue whether this is
really mind blowing. Probably, being my usual excited self, I just said it in
a way that it sounded mind blowing (gesticulating widely and jumping up and
down enthusiastically, you can picture the scene) so my colleague was just
being polite. That was over drinks at the pub after work anyway.
[⁹] Resolution is an inference rule that allows the derivation of new atoms
from a set of clauses. In theorem proving it's used to refute a goal clause by
deriving the empty clause, □. Since a goal is a set of negated literals,
refuting the goal means basically that the negated literals are true. So our
query is true in the context of our program.
[¹⁰] Datalog's bottom-up evaluation uses something called a TP operator. It
basically does what I said above, starts with the ground atoms in a program
and then derives the set of consequences of the clauses in the program. In
each iteration, the set of consequences are added to the program and the
process is repeated, until no new consequences are derived. As stated above,
the process is guaranteed to complete because every datalog definite program
has a least fixpoint, which is also its Least Herbrand Model (we won't go
into Herbrand Models and Herbrand interpretations, but, roughly, an LHM is the
smallest set of atoms that make the union of a definite program and a goal
true). A more complete introduction to LHMs and LFPs and how they are used in
bottom-up evaluation for datalog can be found here:
This is left-recursive, so typical queries will not terminate in Prolog, i.e., have no finite solutions. But as you say, Datalogs are decidable and any query terminates, so you will get solutions, which is different semantics from Prolog. So it's not meaningful to say that Datalog is a semantic subset of Prolog.
Datalog and Prolog are like C++ and Java: One is an extension of a subset of the other, or equivalently, there is a non-empty common subset with similar-ish semantics. This is not a very useful statement? I agree! But it is what it is. They are different languages.
Yes, you're right and I'm also fudging things- but didn't I say that upfront? I start my comment by announcing an "almost not completely fudgy intro to logic programming"!
More seriously, you're right about syntax so thanks for the correction.
But, regarding semantics, the ancestor_of/2 program above can terminate in Prolog, evaluated by SLG resolution, as per my footnote 5. There are still situations where Prolog will not terminate when evaluating a normal program even under SLG resolution, but left recursion is not one of those.
Edit: also, if a Datalog program is also Prolog, and assuming that the program terminates under Prolog, then Prolog and Datalog will both compute its LHM. So it makes sense to say that the two languages are semantically at least very similar and to explain one in terms of the other. They are both much closer than what each is to ASP, for example. It really depends on what assumptions one makes- and that's where the "fudging" comes in.
Anyway, thanks for the correction. I sure could have done a better job of that comment. Did you find anything other that was very wrong in my comment? I'd appreciate it if you pointed it out.
> Did you find anything other that was very wrong in my comment?
I'm not sure about the statement that "Datalog is evaluated from the "bottom up". [...] datalog will first generate the set of all ground atoms that are consequences of the program in the context of which the query was made, then determine whether the atoms in the query are in the set of consequences of the program". I think it's true that Datalog behaves as if it were evaluated like this, but AFAIK Datalog systems can do lots of very aggressive optimizations that change the actual evaluation. I'm not an expert on Datalog.
Anyway, I was mostly dissatisfied with the general thrust of the comment, trying to establish a "subset" relationship. I think the article's "cousins" comment is fair. Prolog is older and influenced Datalog very much. Prolog is also Turing complete, so it is strictly more powerful. But Datalog's semantics allow some optimizations that wouldn't be possible in Prolog, so it can be a lot faster on appropriate classes of problems.
I thought more of your earlier comment about how "p,q:- r" is Datalog. I
accepted this because I figured you know what you're talking about but, to be
honest, I don't know what you're talking about. My understanding is that
without definite clauses (and "p,q:-r" is not definite) there are no fixpoint
semantics and without fixpoint semantics there is no guarantee of program
termination.
So I have to ask: where does this information come from? Could you point me to
a source? To be honest, I suspect that I am confused because of a lack of an
ISO standard for Datalog. Even Prolog, with an ISO standard, has various
extensions like, if memory serves, B-Prolog (which includes OOP elements). If
Datalog has no commonly recognised standard, then basically anything can be
called "Datalog" as long as someone, somewhere, can recognise it as Datalog.
Is that the case here? If so, that would clear my confusion.
It would also make more sense for me to say that "Datalog is a subset of
Prolog", at least in terms of syntax. In that case I'd have to clarify that
I'm talking aout a simple, commonly accepted language fragment of definite
clauses with at most one function symbol, which I think everyone would readily
recognise as "Datalog" without much fuss.
I confess that the extent of my knowledge about Datalog comes from
conversations with (senior) colleagues and not directly from original sources.
Of course, original sources go way, way back (but obviously not as back as
original Prolog sources and I'm familiar with those, so that's not a complete
excuse). In any case, I had a look at, e.g. "What you always wanted to know
about Datalog (and never dared to ask)" by Geri, Gotlob and Tanca (also in the
bibliography section of the wikipedia article on databases).
The article, which is from 1989 and clearly addressed at the databases
community (rather than the logic progamming, or AI community) states that
Datalog is "in many respects a simplified version of general Logic
Programming", referencing J. W. Lloyd for the latter. Since it's 1989, "Logic
Programming" clearly means Prolog (given that it's too early for, e.g. ASP).
Further, the article makes it explicity that "In the formalism of Datalog both
facts and rules are represented as Horn clauses of the general shape ..." and
gives an example of a definite program clause. Later, the article states "From
the syntactic point of view, Datalog is a subset of Prolog" (but then goes on
to point out the difference in semantics).
Finally, the article does agree with you that is is possible to evaluate
Datalog programs in a "top-down" fashion, using the query-subquery algorithm
which is, from what I can tell, backwards chaining implemented by
breadth-first search. So the semantics of Datalog can be different than
Prolog's. My mistake I think is in equating the use of a TP operator (which is
"bottom-up") to Datalog execution, always.
> So I have to ask: where does this information come from?
Half-remembered university courses from >= 10 years ago, I'm afraid. I now think I was probably thinking of ASP, not Datalog. See for example the "Disjunctive logic programs" section of https://www.cs.uni-potsdam.de/~torsten/Lehre/ASP/Folien/asp-.... Since the syntax is very Prolog/Datalog-like, I probably got mixed up. My bad, sorry. The first few sources on Datalog I looked at only talked about Horn clauses, not ones with more heads.
> My understanding is that without definite clauses (and "p,q:-r" is not definite) there are no fixpoint semantics and without fixpoint semantics there is no guarantee of program termination.
Ah, phew, OK. I was really confused by that. No worries, I'm still grateful for your comments :)
Yes, ASP is not Horn so it has rules with multiple literals in the head. "Choice rules", writen as {s,t}:- p. Also, like you say, I believe it doesn't suffer from Prolog's non-termination, again because unlike Prolog it's not Turing-complete. But in any case ASP is based on stable model semantics, not fixpoint semantics.
However I'm really very far from being anything like an expert in ASP! I really should learn a bit of it because it's actually necessary in my research.
Hi, author here. Thank you so much for this unfudging! I knew I was fudging things up a bit, and I didn't really understand the relationships between Prolog, Datalog, Horn clauses, and first-order logic. This clears things up a lot, though I think I'll need to read it a couple more times for it to sink in! Thanks again.
Hey dang, I added some whitespace twice (I needed to add a newline above the first set of Horn clauses to make it into a blockquote) right after I made my comment and the comment "fell" maybe five places. Is that as it should be?
Edit: I did make two or three more changes after adding the whitespace that weren't whitespace, but single characters.
(Not dang!) New comments start at the top, so they have a chance to get upvotes, and quickly drop down if they don't get votes relative to other comments. It sounds normal.
Yes, but a new comment doesn't immediately sink five places, unless the comments above it are very highly upvoted - or the comment is very large. That is, there is a grace period that keeps a comment on top of a thread for a little while, presumably to give it a chance to be read. But, I've noticed that large comments are automatically pushed down into a thread so the grace period for them is much reduced (and probably weighted by the size of the comment).
That's not absolute, because I remember it stated elesewhere by dang that larger comments are considered more likely to be more substantial (for sound reasons). But it seems that holds up to a certain point. There is probably some kind of comment size range, of maximum and minimum comment size, outwith which a comment is pushed down.
There are other factors, also that cause a comment to automatically sink. For example, I have the (admittedly very bad) habbit of editing and re-editing my comments over and over again. I've noticed that if I do this in a sufficiently large comment, the comment immediately sinks when I refresh the thread page after an edit. If the comment is smaller, it can take quite a bit more editing. I think this behaviour is also weighted by user karma or time since account creation because my comments started sinking less often as my karma increased. Or maybe the rules of the site changed in the meantime.
Bottom line: large comments with some edits sink like lead.
With https://github.com/TreeTide/underhood, my goal is to provide a read-only view of code, geared for understanding and debugging. Having to maintain the ability to edit comes with constraints.
Sorry to disappoint but for people doing programming language semantics there’s nothing new here. The hard part is always to go from the made-up language to a set of multiple full-blown languages.
> I want to see how easy it is to add more advanced features to FP, like generics
This will be interesting to see, since the straight-forward implementation of generics for functional programming languages uses unification, which is not available in Datalog. It will probably be possible to encode things for any given program, since its universe of types should be finite. But it will involve jumping through hoops to encode something equivalent to unification.
I was very confused as I had read the title as '... with Datadog" (the monitoring tool).
Anyway, I think both are a good idea. Datadog -> integrating monitoring information back into the IDE. Datadog -> working with the actual semantics of the code rather than just blobs of text.
I know there's more to just queriability in the post, but I thought this should've been mentioned when discussing what existing IDEs can offer.