Damn you smalltalk⦠you suck me in to little side problemsâ¦. RBParser is actually a lot more flexible than I had realised - and not so difficult to understand for a mere mortal⦠It naively seems that (what I still consider a confusingly named method) is quite simple to extend to handle at least medium sized errors just by doing something like: ast := [RBParser parseMethod: source onError: [ :msg :pos :parser | |seqNode errorNode| errorNode := RBParseErrorNode errorMessage: msg value: parser currentToken value printString at: pos. seqNode := parser sequenceNodeClass new. parser step; parseStatements: true into: seqNode. seqNode addNodeFirst: errorNode; yourself ]] onDNU: #body: do: [ nil âHandle non method source fragments"]. NOTE: I did have to add >>#currentToken as a method to RBParser as it doesnât expose very much for the onError blocks, which I think it should to make them a bit more useful. Tim
On 7 Sep 2017, at 14:39, Tim Mackinnon <tim@testit.works> wrote:
Thanks Thierry - Iâve been noodling in the background and next I hit the question of âcommentsâ - which I found RB does handle⦠and then as per your point - how to skip past an error and try further down (which it seems that the scanner might have the info to let me skip simple things). You seem to get sucked in more and more when trying to to this properly - which does make it sound like a very good internship or GSoc project.
I think I may have enough to get something polished enough to release to the wider world. I certainly love having better keyboard support when Iâm coding - as we type a lot (even though we have simple syntax) - and having efficient tools takes away the friction.
Next on my hit list, it add the same support to the refactoring tools - which while sophisticated are so clunky in use when compared to IntelliJ.
Tim
On 7 Sep 2017, at 08:02, Thierry Goubier <thierry.goubier@gmail.com <mailto:thierry.goubier@gmail.com>> wrote:
Hi Tim,
2017-09-01 13:39 GMT+02:00 Tim Mackinnon <tim@testit.works <mailto:tim@testit.works>>: Thanks Thierry - this is proving an interesting problem domain ...
I too am using #bestNodeFor: although I don't find it always gives the best node (it does if you are clearly in the target range, but if you are on the outer boundary like the next position after a selector or next to a "." or ";" it favours the parent and not the closest node. So in usage I find I need to tweak it a bit.
Yes, when you are on a boundary, then finding the right node may be difficult. When developping the ancestor to smart suggestion, I considered #bestNodeFor: as good enough, but I considered looking into "up" and "down" ast navigation at a time, and I wasn't alone (I think it was active in the Pharo editor at a point).
I'll look at smacc though - also there is that experimental setting to allow parse errors, I don't know if it helps in any way.
There is a general question there, which is how you try to parse as much as possible while jumping over the error, making the error area as small as possible (that would be really great for syntax highlighting, by the way).
The problem is that in hand-written parsers, you need to hard-code the error management in the parser. With a table-driven parser (as is SmaCC), then you can explore in a systematic way what are the possible states that would allow the parsing to continue correctly after making the error area as small as possible.
This would make for a very nice internship subject...
All this said, I think I have a workable suggestion that is not much code that might be open to scrutiny from those wiser than me.
Looking at the keyboard handling in the system - it's quite sprawling and tricky to follow. I think there is lots of room for refactoring.
It is very easy in Pharo to implement a nice and clean keyboard handling; all the necessary components have been in place for years, and we already have implementations (using the KM API).
Now, most Pharo developpers just jump into hardcoding keyboard handling instead.
Thierry
I'm also not sure if I like the pragma usage either - I find it quite difficult to follow what's calling what.
Tim
Sent from my iPhone
On 1 Sep 2017, at 09:18, Thierry Goubier <thierry.goubier@gmail.com <mailto:thierry.goubier@gmail.com>> wrote:
Hi Tim,
The RB ast has a specific method for finding the right node: in AltBrowser, I use ast bestNodeFor: aTarget selectionInterval.
For the second case (parse what is selected), you may want to look into SmaCC: it has specific entry points (used for refactoring) to try all possible starts in a parser for a given piece of text and returning all the possible correct asts. As a hand-writen parser, the RBParser can't do that unless you add the necessary entry points by hand and hack around the error handling, because most of the parse attempts end on an error (as they should).
Regards,
Thierry
2017-09-01 8:50 GMT+02:00 Tim Mackinnon <tim@testit.works <mailto:tim@testit.works>>: Marcus - I'd be interested in how you thought to solve this.
My solution seems to work - but it's not the most elegant. Retrying lots of different ways like I show, smacks a bit of desperation (this said, it is quite compact).
Apart from that - I was actually interested in thoughts on the parse method - the onError naming is a bit misleading compared to other methods in the image. I wonder if it should be called something like "onErrorMerge:" to signal that it's going to reuse the results?
Stef - with regards to broken source, my proposed solution reuses what is already there - thus if you highlight specific code, it reverts to the original strategy and tries to compile just that code. This bugs me however - so if you highlight nothing, it tries to compile the source multiple ways, I drop down to chopping the source up to where your cursor is - which seems like a decent cheap strategy.
I don't know if our parser is able to recover from simple errors and just have error nodes in the ast, and whether the #bestNodeFor method then ignored error nodes... this is what I think Dolphin used to do.
Tim
Sent from my iPhone
On 31 Aug 2017, at 19:11, Marcus Denker <marcus.denker@inria.fr <mailto:marcus.denker@inria.fr>> wrote:
On 31 Aug 2017, at 19:07, Stephane Ducasse <stepharo.self@gmail.com <mailto:stepharo.self@gmail.com>> wrote:
On Wed, Aug 30, 2017 at 9:50 PM, Tim Mackinnon <tim@testit.works <mailto:tim@testit.works>> wrote:
Iâm looking for some feedback on the usage of RBParser>>#parseMethod:onError:.
I am looking at ways of improving the way we edit code - and making things work (as least for me - but maybe for everyone) a bit more like IntelliJ where they have some great code operations and very good keyboard support. We are certainly close, but at times feel a bit clumsy - which is a shame as we have far better infrastructure to support equivalent features. So I thought I would have a go.
Excellent!!!
Anyway - step one (which I think Iâve come close on), was to teach senders/implementors to look at the AST so you donât have to highlight exactly what you want to search on - instead it can infer from your cursor⦠however my next step was to improve how we can select code to ease refactoring, bracketing etcâ¦
Yes but we have to handle broken code then. Did you check how smart suggestions did it for code that compiled?
I looked some weeks ago and made some notes what to do⦠but I fear this has to wait for next week, there is no way that I can do anything till ESUGâ¦
Marcus