Issue #20024 has been updated by kddnewton (Kevin Newton). I understand the need for compatibility with regard to Prism, but this ticket isn't about Prism specifically. I'm proposing this as a solution to the problem in general, devoid of the Prism context. Regardless of other parsers, there exists a problem that tools are relying on error messages for behavior. We can fix that by introducing subclasses. Of course it won't fix the immediate situation for anyone, but going forward it will be a much better situation. ---------------------------------------- Feature #20024: SyntaxError subclasses https://bugs.ruby-lang.org/issues/20024#change-105436 * Author: kddnewton (Kevin Newton) * Status: Open * Priority: Normal ---------------------------------------- There are many places around the Ruby ecosystem that handle syntax errors in different ways. Some provide highlighting, others provide recovery of some form, still more provide LSP metadata. In order to provide more rich information, most of them switch on the message of the error being returned, as in: https://github.com/ruby/irb/blob/f86d9dbe2fc05ed62332069a27f4aacc59ba9634/li... Within ruby/spec, specific error messages are required for these kinds of messages in order to support this implicit interface that syntax errors have a hidden type, which is only expressed through their message. For example: https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... It's not clear from these specs or from the parser itself which error messages are permanent/guaranteed versus which are changeable. Either way, relying on the error message itself as opposed to the type of the error is brittle at best. I would like to suggest instead we implement subclasses on `SyntaxError` that would allow tools that depend on specific syntax errors to rescue those subclasses instead of parsing the message. In addition to alleviating the need to parse error messages with regex, this would also allow for the possibility that the error messages could change in the future without breaking external tooling. Allowing these to change would allow them to be potentially enhanced or changed by other tools - for example by providing recovery information or translating them. This is particularly important for Prism since we are getting down to individual spec failures and some of the failures are related to the fact that we have messages like `"Numbered parameter is already used in outer scope"` where the spec requires `/numbered parameter is already used in/`. Even this case-sensitivity is causing failures, which seems like we're testing the wrong thing. -- https://bugs.ruby-lang.org/

Issue #20024 has been updated by Eregon (Benoit Daloze). A Symbol per type of SyntaxError seems easier to maintain and wouldn't create so many classes/constants/etc which would be messy in docs. I think that's a much easier improvement to do than adding a bunch of subclasses. It's also better memory-wise as classes are not small. ---------------------------------------- Feature #20024: SyntaxError subclasses https://bugs.ruby-lang.org/issues/20024#change-105438 * Author: kddnewton (Kevin Newton) * Status: Open * Priority: Normal ---------------------------------------- There are many places around the Ruby ecosystem that handle syntax errors in different ways. Some provide highlighting, others provide recovery of some form, still more provide LSP metadata. In order to provide more rich information, most of them switch on the message of the error being returned, as in: https://github.com/ruby/irb/blob/f86d9dbe2fc05ed62332069a27f4aacc59ba9634/li... Within ruby/spec, specific error messages are required for these kinds of messages in order to support this implicit interface that syntax errors have a hidden type, which is only expressed through their message. For example: https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... It's not clear from these specs or from the parser itself which error messages are permanent/guaranteed versus which are changeable. Either way, relying on the error message itself as opposed to the type of the error is brittle at best. I would like to suggest instead we implement subclasses on `SyntaxError` that would allow tools that depend on specific syntax errors to rescue those subclasses instead of parsing the message. In addition to alleviating the need to parse error messages with regex, this would also allow for the possibility that the error messages could change in the future without breaking external tooling. Allowing these to change would allow them to be potentially enhanced or changed by other tools - for example by providing recovery information or translating them. This is particularly important for Prism since we are getting down to individual spec failures and some of the failures are related to the fact that we have messages like `"Numbered parameter is already used in outer scope"` where the spec requires `/numbered parameter is already used in/`. Even this case-sensitivity is causing failures, which seems like we're testing the wrong thing. -- https://bugs.ruby-lang.org/

Issue #20024 has been updated by kddnewton (Kevin Newton). I think I am okay with that compromise. I don't like that you won't be able to rescue specific errors, but I understand the concern about forward compatibility. My other desire for subclasses is that different syntax errors have different information on them. For example, `_1 = 1` is a syntax error that has the additional information that you're assigning to `_1`. Most libraries right now parse that out of the message, but if we had separate objects then we could define the shapes of the errors to have that information as a field. For example, that would be the `name` or `identifier` field. This could probably be addressed with something like a `metadata` hash or something else, if we're just adding fields to SyntaxError. ---------------------------------------- Feature #20024: SyntaxError subclasses https://bugs.ruby-lang.org/issues/20024#change-105440 * Author: kddnewton (Kevin Newton) * Status: Open * Priority: Normal ---------------------------------------- There are many places around the Ruby ecosystem that handle syntax errors in different ways. Some provide highlighting, others provide recovery of some form, still more provide LSP metadata. In order to provide more rich information, most of them switch on the message of the error being returned, as in: https://github.com/ruby/irb/blob/f86d9dbe2fc05ed62332069a27f4aacc59ba9634/li... Within ruby/spec, specific error messages are required for these kinds of messages in order to support this implicit interface that syntax errors have a hidden type, which is only expressed through their message. For example: https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... It's not clear from these specs or from the parser itself which error messages are permanent/guaranteed versus which are changeable. Either way, relying on the error message itself as opposed to the type of the error is brittle at best. I would like to suggest instead we implement subclasses on `SyntaxError` that would allow tools that depend on specific syntax errors to rescue those subclasses instead of parsing the message. In addition to alleviating the need to parse error messages with regex, this would also allow for the possibility that the error messages could change in the future without breaking external tooling. Allowing these to change would allow them to be potentially enhanced or changed by other tools - for example by providing recovery information or translating them. This is particularly important for Prism since we are getting down to individual spec failures and some of the failures are related to the fact that we have messages like `"Numbered parameter is already used in outer scope"` where the spec requires `/numbered parameter is already used in/`. Even this case-sensitivity is causing failures, which seems like we're testing the wrong thing. -- https://bugs.ruby-lang.org/

Issue #20024 has been updated by jeremyevans0 (Jeremy Evans). zverok (Victor Shepelev) wrote in #note-7:

An aside note on this: in general, `rescue` clause is already looking very alike pattern-matching, and if we continue this approach (with adding causes/details to exceptions), we might once want to be able to...

```ruby rescue SyntaxError(issue: :void_value) => ex ```

That's already valid syntax, which calls the `SyntaxError` method with an `issue` keyword argument (the method should return a class or module to work correctly). ---------------------------------------- Feature #20024: SyntaxError subclasses https://bugs.ruby-lang.org/issues/20024#change-105442 * Author: kddnewton (Kevin Newton) * Status: Open * Priority: Normal ---------------------------------------- There are many places around the Ruby ecosystem that handle syntax errors in different ways. Some provide highlighting, others provide recovery of some form, still more provide LSP metadata. In order to provide more rich information, most of them switch on the message of the error being returned, as in: https://github.com/ruby/irb/blob/f86d9dbe2fc05ed62332069a27f4aacc59ba9634/li... Within ruby/spec, specific error messages are required for these kinds of messages in order to support this implicit interface that syntax errors have a hidden type, which is only expressed through their message. For example: https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... It's not clear from these specs or from the parser itself which error messages are permanent/guaranteed versus which are changeable. Either way, relying on the error message itself as opposed to the type of the error is brittle at best. I would like to suggest instead we implement subclasses on `SyntaxError` that would allow tools that depend on specific syntax errors to rescue those subclasses instead of parsing the message. In addition to alleviating the need to parse error messages with regex, this would also allow for the possibility that the error messages could change in the future without breaking external tooling. Allowing these to change would allow them to be potentially enhanced or changed by other tools - for example by providing recovery information or translating them. This is particularly important for Prism since we are getting down to individual spec failures and some of the failures are related to the fact that we have messages like `"Numbered parameter is already used in outer scope"` where the spec requires `/numbered parameter is already used in/`. Even this case-sensitivity is causing failures, which seems like we're testing the wrong thing. -- https://bugs.ruby-lang.org/

Issue #20024 has been updated by zverok (Victor Shepelev). @jeremyevans0

That's already valid syntax

I am aware; I just have a feeling that repurposing it with PM might still be possible (the chance that a lot of code uses this syntax with `UpcasedMethods` to produce exception classes dynamically is relatively low—though, I admit, not non-existent). Anyway, I think @kddnewton is correct: here is not the best place, and the idea is very rough, I am sorry for stealing the discussion, just got a sudden idea. ---------------------------------------- Feature #20024: SyntaxError subclasses https://bugs.ruby-lang.org/issues/20024#change-105445 * Author: kddnewton (Kevin Newton) * Status: Open * Priority: Normal ---------------------------------------- There are many places around the Ruby ecosystem that handle syntax errors in different ways. Some provide highlighting, others provide recovery of some form, still more provide LSP metadata. In order to provide more rich information, most of them switch on the message of the error being returned, as in: https://github.com/ruby/irb/blob/f86d9dbe2fc05ed62332069a27f4aacc59ba9634/li... Within ruby/spec, specific error messages are required for these kinds of messages in order to support this implicit interface that syntax errors have a hidden type, which is only expressed through their message. For example: https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... It's not clear from these specs or from the parser itself which error messages are permanent/guaranteed versus which are changeable. Either way, relying on the error message itself as opposed to the type of the error is brittle at best. I would like to suggest instead we implement subclasses on `SyntaxError` that would allow tools that depend on specific syntax errors to rescue those subclasses instead of parsing the message. In addition to alleviating the need to parse error messages with regex, this would also allow for the possibility that the error messages could change in the future without breaking external tooling. Allowing these to change would allow them to be potentially enhanced or changed by other tools - for example by providing recovery information or translating them. This is particularly important for Prism since we are getting down to individual spec failures and some of the failures are related to the fact that we have messages like `"Numbered parameter is already used in outer scope"` where the spec requires `/numbered parameter is already used in/`. Even this case-sensitivity is causing failures, which seems like we're testing the wrong thing. -- https://bugs.ruby-lang.org/

Issue #20024 has been updated by yui-knk (Kaneko Yuichiro). # SyntaxError includes multiple errors `SyntaxError` includes multiple errors like below, in this case two errors are included into one `SyntaxError`. Therefore it needs to consider how to handle such cases. ```ruby begin eval <<~CODE def m retry CODE rescue SyntaxError => e puts e.message end # (eval at test.rb:2):2: Invalid retry without rescue # retry # ^~~~~ # (eval at test.rb:2):3: syntax error, unexpected end-of-input, expecting `end' or dummy end ``` We need to avoid losing information to provide rich information to `SyntaxError` users. Therefore these are not options for this problem: * Merge multiple errors into one `SyntaxError` subclass. Because this is misleading if these errors are different types of errors. * Use only a single error and ignore following erorrs. Because information is lost. # irb use case and error tolerance parser Regarding the [irb use case](https://github.com/ruby/irb/blob/f86d9dbe2fc05ed62332069a27f4aacc59ba9634/li...), it categorizes error , which is recoverable by adding tokens to the end, as `recoverable_error` and error, which is recoverable by deleting tokens, as `unrecoverable_error`, so that irb can determine to require more input or not. When irb was created, error tolerance parser didn't exist. Therefore irb needs to take care of how to recover the input using `SyntaxError#message`. However it's parser responsibility to recover errors. irb can categorize syntax error if error tolerance parser provides information how the parser recover errors. If only single error and recovery requires only token insert operations for the last of input, it's `recoverable_error`. By the way, does the parser always raise `SyntaxError` for invalid input? For this use case, it's better to not raise `SyntaxError` but users ask to parser to have syntax errors or not after parsing. # Proposal Considering these points, my proposal is defining parser interface. * Parser can run without SyntaxError mode * Parser provides a method to get syntax errors * Each syntax error includes * message: "syntax error, unexpected end-of-input, expecting `end' or dummy end" * location: (1,0)-(1,1), first/last & line/column * operations for recovery: * insert / delete * location of recovery ---------------------------------------- Feature #20024: SyntaxError subclasses https://bugs.ruby-lang.org/issues/20024#change-105468 * Author: kddnewton (Kevin Newton) * Status: Open * Priority: Normal ---------------------------------------- There are many places around the Ruby ecosystem that handle syntax errors in different ways. Some provide highlighting, others provide recovery of some form, still more provide LSP metadata. In order to provide more rich information, most of them switch on the message of the error being returned, as in: https://github.com/ruby/irb/blob/f86d9dbe2fc05ed62332069a27f4aacc59ba9634/li... Within ruby/spec, specific error messages are required for these kinds of messages in order to support this implicit interface that syntax errors have a hidden type, which is only expressed through their message. For example: https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... It's not clear from these specs or from the parser itself which error messages are permanent/guaranteed versus which are changeable. Either way, relying on the error message itself as opposed to the type of the error is brittle at best. I would like to suggest instead we implement subclasses on `SyntaxError` that would allow tools that depend on specific syntax errors to rescue those subclasses instead of parsing the message. In addition to alleviating the need to parse error messages with regex, this would also allow for the possibility that the error messages could change in the future without breaking external tooling. Allowing these to change would allow them to be potentially enhanced or changed by other tools - for example by providing recovery information or translating them. This is particularly important for Prism since we are getting down to individual spec failures and some of the failures are related to the fact that we have messages like `"Numbered parameter is already used in outer scope"` where the spec requires `/numbered parameter is already used in/`. Even this case-sensitivity is causing failures, which seems like we're testing the wrong thing. -- https://bugs.ruby-lang.org/

Issue #20024 has been updated by kddnewton (Kevin Newton). @yui-knk your proposal matches how prism works today. Prism never raises a syntax error, it returns a list of errors that have (exactly as you have said) a type, a message, and a location. I agree that a single raised error is not great because it loses information. I like the idea that @ioquatix has proposed of `SyntaxError#diagnostics`. With such a proposal, we could attach all syntax errors in the file to a single raised error, and you could query them appropriately. In this case we would need to design the shape of a `SyntaxError::Diagnostic` class. I would imagine it would have the same fields that @yui-knk is proposing, something like: ```ruby class SyntaxError::Location attr_reader :start_line, :end_line # 1-indexed attr_reader :start_column, :end_column # in bytes end class SyntaxError::Diagnostic attr_reader :location # SyntaxError::Location attr_reader :message # String attr_reader :type # Symbol end class SyntaxError attr_reader :diagnostics # Array[SyntaxError::Diagnostic] end ``` Would that be amenable? ---------------------------------------- Feature #20024: SyntaxError subclasses https://bugs.ruby-lang.org/issues/20024#change-105519 * Author: kddnewton (Kevin Newton) * Status: Open * Priority: Normal ---------------------------------------- There are many places around the Ruby ecosystem that handle syntax errors in different ways. Some provide highlighting, others provide recovery of some form, still more provide LSP metadata. In order to provide more rich information, most of them switch on the message of the error being returned, as in: https://github.com/ruby/irb/blob/f86d9dbe2fc05ed62332069a27f4aacc59ba9634/li... Within ruby/spec, specific error messages are required for these kinds of messages in order to support this implicit interface that syntax errors have a hidden type, which is only expressed through their message. For example: https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... It's not clear from these specs or from the parser itself which error messages are permanent/guaranteed versus which are changeable. Either way, relying on the error message itself as opposed to the type of the error is brittle at best. I would like to suggest instead we implement subclasses on `SyntaxError` that would allow tools that depend on specific syntax errors to rescue those subclasses instead of parsing the message. In addition to alleviating the need to parse error messages with regex, this would also allow for the possibility that the error messages could change in the future without breaking external tooling. Allowing these to change would allow them to be potentially enhanced or changed by other tools - for example by providing recovery information or translating them. This is particularly important for Prism since we are getting down to individual spec failures and some of the failures are related to the fact that we have messages like `"Numbered parameter is already used in outer scope"` where the spec requires `/numbered parameter is already used in/`. Even this case-sensitivity is causing failures, which seems like we're testing the wrong thing. -- https://bugs.ruby-lang.org/

Issue #20024 has been updated by palkan (Vladimir Dementyev). kddnewton (Kevin Newton) wrote in #note-14:

In this case we would need to design the shape of a `SyntaxError::Diagnostic` class. I would imagine it would have the same fields that @yui-knk is proposing, something like: ...

I think, we can get some inspiration from Parser: [diagnostic](https://github.com/ruby-next/parser/blob/master/lib/parser/diagnostic.rb) ([messages](https://github.com/ruby-next/parser/blob/master/lib/parser/messages.rb) — a lot of them, btw). For example, it also includes a _level_ information (error, fatal, warn); something similar to the "recoverable" property proposed above, but more granular. ---------------------------------------- Feature #20024: SyntaxError subclasses https://bugs.ruby-lang.org/issues/20024#change-106168 * Author: kddnewton (Kevin Newton) * Status: Open * Priority: Normal ---------------------------------------- There are many places around the Ruby ecosystem that handle syntax errors in different ways. Some provide highlighting, others provide recovery of some form, still more provide LSP metadata. In order to provide more rich information, most of them switch on the message of the error being returned, as in: https://github.com/ruby/irb/blob/f86d9dbe2fc05ed62332069a27f4aacc59ba9634/li... Within ruby/spec, specific error messages are required for these kinds of messages in order to support this implicit interface that syntax errors have a hidden type, which is only expressed through their message. For example: https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... https://github.com/ruby/spec/blob/c3206f644325c026fc5b700f0ea75ce9bd2e9d02/l... It's not clear from these specs or from the parser itself which error messages are permanent/guaranteed versus which are changeable. Either way, relying on the error message itself as opposed to the type of the error is brittle at best. I would like to suggest instead we implement subclasses on `SyntaxError` that would allow tools that depend on specific syntax errors to rescue those subclasses instead of parsing the message. In addition to alleviating the need to parse error messages with regex, this would also allow for the possibility that the error messages could change in the future without breaking external tooling. Allowing these to change would allow them to be potentially enhanced or changed by other tools - for example by providing recovery information or translating them. This is particularly important for Prism since we are getting down to individual spec failures and some of the failures are related to the fact that we have messages like `"Numbered parameter is already used in outer scope"` where the spec requires `/numbered parameter is already used in/`. Even this case-sensitivity is causing failures, which seems like we're testing the wrong thing. -- https://bugs.ruby-lang.org/