Hi Daniel. Quick question: Does this use the same underlying parse nodes, and helper
classes like ParseTreeRewriter, as the old system, just generating them a different way? If so, I think the value here is as much in the convenience methods for performing searches and rewrites as the pattern syntaxâso thank you for that, regardless!
Yes, underhood BPatterns is based on the pattern AST nodes and the matching/rewrite machinery. It's just a different DSL to the original engine. There are some changes to better follow my idea but it is not significant. While we're on the subject of rewrite tools, a while back I noticed that I
often wanted to rewrite sequences of statements to a single method call, where that sequence might recur several times in the same method. This especially comes up in testsâperhaps I notice that the pattern:
self assert: obj foo equals: `#expectedFoo.
self assert: obj bar equals: `#expectedBar. self assert: obj baz equals: `#expectedBaz.
Agree, I have the same feeling about the sequences. And it can be quite confusing when you do not know the internal details. I don't think I will have any time to change this. вÑ, 3 ÑевÑ. 2026â¯Ð³. в 23:51, Daniel Slomovits <daniels220@gmail.com>:
I love the basic idea, but at a glance, it seems like the syntax rapidly becomes *much* more verbose than the equivalent using the special pattern syntax. For simple examples it's more-or-less equivalent, but the with: [ anyVar ] -> [:pattern | ...] syntax is *very* clunky.
What I would *really* love to see is an extension to the Compiler, that allows for mixing the actual pattern syntax directly into code. Perhaps initiated by a backtick-parenthesis, so you could write:
(`(``@receiver isNil ifTrue: ``@nilBlock) -> `(``@receiver ifNil: ``@nilBlock)) rewrite preview
I'm not even sure this would be all that hardâit would mean adding one new type of parse node, call it a "pattern literal", delimited by `(), whose implementation would be very straightforward for the most part, with the only complication being that pattern blocks (`{}) would have to be compiled properly in the outer scope. As additional niceties, individual pattern nodes should ideally be able to appear bare, rather than needing to be wrapped in `(), and we would want a syntax for "this is a whole-method pattern"
On another note, I want to call attention to one thing that BPattern has going for it that string-literal-based patterns can't (easily) do: modifier blocks are closures in the method that generated them, so you can do things like have methods that return a pattern which closes over the arguments of the method...having trouble coming up with an example off the top of my head, but this is very powerful. (Well. Actually you *can* do this with string-based patterns, by reaching in after the pattern is parsed and editing the modifier block. I've written methods to support the following syntax: '`@exp' withPatternBlock: [:exp | "this is a real block, although the expression it wraps was still parsed from a string literal"].)
Quick question: Does this use the same underlying parse nodes, and helper classes like ParseTreeRewriter, as the old system, just generating them a different way? If so, I think the value here is as much in the convenience methods for performing searches and rewrites as the pattern syntaxâso thank you for that, regardless!
While we're on the subject of rewrite tools, a while back I noticed that I often wanted to rewrite sequences of statements to a single method call, where that sequence might recur several times in the same method. This especially comes up in testsâperhaps I notice that the pattern:
self assert: obj foo equals: `#expectedFoo. self assert: obj bar equals: `#expectedBar. self assert: obj baz equals: `#expectedBaz.
occurs a lot, and those properties are related enough that it makes more sense to say self verifyFooBarBaz: obj equal: {`#expectedFoo. `#expectedBar. `#expectedBaz}. (This is a poor example, in truth, but there are cases where it's much more clearly justified.) By default you can only get one of these occurrences per run of the rewrite tool, because any pattern you write has to match the *entire sequence*, and once it has matched, the entire sequence is disqualified from further matching. I was able to extend the matching semantics so that a recurse-into-list-of-statements pattern (i.e. ``@.stmts, double-backtick-at-period) effectively builds a *new* sequence containing only the statements it matched, and retries the rule against *that*. It could definitely stand to be more elegant, but for me it filled an important hole in the existing capabilities. Is that something anyone would be interested in helping me dust off and contribute?
On Tue, Feb 3, 2026 at 1:41â¯PM Denis Kudriashov <dionisiydk@gmail.com> wrote:
Hi Richard.
I checked the code and the double backticks are for the recursion mode. It forces the search to dive into the current matching node to find more cases inside it. For example:
(instVar isNil ifTrue: [ 'inside recursion' ]) isNil ifTrue: [ 'outer node' ]
The syntax with single backtick will not find the internal "instVar isNil" expression.
In BPatterns the recursion mode is enabled by default for the pattern variables. So the default mode covers the maximum scope for a search. Currently you can't not disable the #recurseInto flag as there is not suitable accessor for this. But if it would be there the pattern would like:
[ anyVar isNil ifTrue: anyBlock ] bpattern with: [ anyVar ] -> [:pattern | pattern recurseInto: false ]
Anyway I adjusted my blog accordingly. Thanks for the attention.
вÑ, 3 ÑевÑ. 2026â¯Ð³. в 10:47, Richard O'Keefe <raoknz@gmail.com>:
Thank you. One of the reasons given for introducing BPatterns is a subtle difference between single backticks and double backticks, which is mentioned but not explained in the BPatterns blog entry. The help screen you directed me to does not explain that.
Is there any documentation which is *complete* and *proofread*?
On Tue, 3 Feb 2026 at 23:16, Denis Kudriashov <dionisiydk@gmail.com> wrote:
Yes. There is nice help presenter in the Pharo:
StRewriterHelpBrowserPresenter new open
[image: Screenshot 2026-02-03 at 10.15.45.png]
вÑ, 3 ÑевÑ. 2026â¯Ð³. в 01:36, Richard O'Keefe <raoknz@gmail.com>:
Is the "classic" rewriting syntax documented anywhere?
On Mon, 2 Feb 2026 at 22:52, Denis Kudriashov via Pharo-users < pharo-users@lists.pharo.org> wrote:
Hi all.
This is the release of *BPatterns* project, one of my old ideas on how to make the rewrite engine usable for scripting.
As a quick reminder, here is an example of the classic rewrite-engine matching syntax:
*``@receiver isNil ifTrue: ``@nilBlock*
In practice, patterns like this usually live inside string literalsâhidden away in lint rules, deprecations, or refactorings. To use them for *manual* code search or rewriting, you typically need a dedicated tool.
Now compare that with the same pattern expressed as a *BPattern*:
*[ any isNil ifTrue: anyBlock ] bpattern*
No special syntax. No magical tools. Just pure Smalltalk.
You write a block. You send #bpattern. Youâre done.
Every editor understands it. Every tool respects it.
Want to see where this pattern appears?
* [ any isNil ifTrue: anyBlock ] bpattern browseUsers*
[image: Screenshot 2026-01-31 at 12.16.19.png]
Want to rewrite all of them?
*[[ any isNil ifTrue: anyBlock ] -> [ any ifNil: anyBlock ]] brewrite preview*
[image: Screenshot 2026-01-31 at 12.17.57.png]
Because *BPatterns are real Smalltalk code*, all development tools work out of the box: syntax highlighting, completion, navigation, refactorings.
This is the rewrite engine treated as a *first-class citizen*.
Not strings. Not tooling sidecars. Not just for advanced users.
*A rewrite engine you can actually use.* * A rewrite engine at your fingertips.* More details are in the blog post:
- https://dionisiydk.blogspot.com/2026/02/bpatterns-rewrite-engine-with-smallt...
And follow the project on GitHub:
- https://github.com/dionisiydk/BPatterns
-
