An Introduction to Condition Handling for Non-Lispers

http://lubutu.com/soso/condition-handling-for-non-lispers

by lubutu • 14 years ago

104 65 14 years ago

qznc5• 14 years ago

What I do not like about this: You expose implementation details to the outside with those recover points. In the same sense exceptions in Java expose implementation details, though.

For this specific example, the better behavior (imho) would be that parseEntry always returns. So it returns ParseError instead of throwing it.

Basically, I want a more convincing example for condition handling.

lubutu3• 14 years ago

Suppose we were attempting to write parseLogInteractively by just returning errors (as in idiomatic Go).

If parseLog just returned an error upon failure we would have to same problem as exceptions do, as explained in section 1. So we have to make parseLog return two lists: one of entries, and one of errors. We would then have to search through the error list fixing and reparsing each error. In Python this would be something like,

  def parseLogInteractively(file):
      return parseLinesInteractively(file.readLines())

  def parseLinesInteractively(lines):
      entries, errors = parseLines(lines)
      if errors:
          entries += parseLinesInteractively(askToFixEntry(e.text) for e in errors)
      return entries

Again we lose the ability to abstract the function 'parseLog'. And I don't know about you, but this looks far worse to me than "resume FixEntry" in terms of exposing implementation details.

oddthink2• 14 years ago

In python, I usually deal with this with generators. It's not as elegant as conditions, but it works well enough. I've never tried to permit custom restart-like behavior, but now that generators are coroutines, it should be doable.

e.g.

    def parseEntry(line):
        try:
            parsed = foo(line)
            yield parsed
        except SomethingBad:
            yield None

Common lisp could really use some better coroutine support (and some way to specify generic sequences). Sure, you can do it with macros, but it gets horrible fast. Have you looked at SERIES, for example?

qznc5• 14 years ago

Do not separate errors and entries. Instead use something like Haskell's Maybe, but instead of an empty Nothing-object you return something which contains error information:

  def parseLogInteractively(file):
    entries = list()
    for line in file:
      entry = parseEntry(line)
      if entry.failed:
        entry = askToFixEntry(entry)
      entries.append(entry)
    return entries

A design question arises: Do you need to fix parse errors before you can proceed with next entry? If not, the error handling can be done by whomever calls parseLog.

lmm3• 14 years ago

Sounds like this is just an opaque way of providing a strategy or callback to the parseLog method. Why not something along the lines of (scala):

  def parseLogInteractively(file) = parseLines(file.readLines(), error => askToFixEntry(e.txt))

  def parseLines(lines, errorHandler) = lines flatMap {
    tryParseLine(_) fold {success(l) => Some(l)} {error => errorHandler(error)}
  }

lispm3• 14 years ago

That's one idea of the Lisp condition system: you don't return. The handler gets called on error and the full context is still there. The context includes all bindings, dynamic handlers and all possible restarts. The restarts itself are most often generic: Abort, Continue, Use Value, Store Value, ... There can also be several of those. For example the log error could be restarted by using a Abort error to abort a specific action (say, a compilation of a file) , to abort the whole process (say, a compilation of a whole system).

qznc5• 14 years ago

My favored method: return an error object, which includes all the necessary context. Of course, the library writer must anticipate what information might be necessary, but with condition handling he also has to anticipate, what restarts are necessary.

astine• 14 years ago

How would you accomplish somthing similar without exposing implementation details? It seems to me "this function may throw these conditions" is more a documentation issue. I think every other system I've seen, (esp C) exposes more details.

draven• 14 years ago

Letting the caller do the handling and giving it access to the dynamic environment where the condition was signaled is a feature.

This paper by Pitman helps to understand the design choices www.nhplace.com/kent/Papers/Condition-Handling-2001.html

masklinn10• 14 years ago

> You expose implementation details to the outside with those recover points.

Of course not, no more than an exception or an error code "exposes implementation details". Condition restarts are part of a library's API, since they're a way for library users to interact with the library (in this case, to customize the handling of some error conditions)

> For this specific example, the better behavior (imho) would be that parseEntry always returns. So it returns ParseError instead of throwing it.

How and why is it a better behavior, especially if — unlike this contrived example — there are 5 frames of foreign code between the caller and the condition being signalled? Return or exception will have to unwind those frames, the point of a condition is that you don't, instead right there the library can provide a hook through which the library user can respond to the question: "how do I handle [this error]?"

Jach• 14 years ago

Controversial programming opinion of the day: not knowing implementation details will come back to bite you, overexposure is better than underexposure.

Anyway, I think the basic idea is just a small extension of being able to pause the program counter, change stuff, do something else, and at one's option continue where one left off, all without requiring a debugger. Nothing new if you've been exposed to the Lisp world. Typical examples of utility being recovering from longComputation() then failing without repeating work and hot-fixing a long-running server process. I agree this example isn't a very good one, I like your design you commented on for this problem better, I think in this case even "an error in parsing a line is not exceptional" would suffice in defeating the example. (Though in Python even syntax errors are implemented with exceptions, so...)

erikb2• 14 years ago

I think you are looking for the "with" statement! The feature you describe is already there. And it works with objects, so it's reusable, extendable and exchangeable if needed.

http://effbot.org/zone/python-with-statement.htm

masklinn10• 14 years ago

No, context management is only a small subset of what you can do with condition (and does not involve restart), and also implies stack unwinding (as do exceptions) amongst other things.

To TFAA: to talk about conditions but not use lisp syntax, you could have used Smalltalk (the other language with conditions and restarts)

erikb2• 14 years ago

We are not talking "condition" from Lisp here, but the blog post that is linked to this thread. And what the author is doing is asking for a "with" statement without knowing that the feature is already there. "condition" might be a powerful feature and I might even like it myself, but the post doesn't show the power of "condition" to Pythoners, because the feature used in the example already exists in the core language.

howeyc• 14 years ago

For anyone that had trouble learning the condition system, or just wants another example, this helped me:

http://symbo1ics.com/blog/?p=1405

qznc5• 14 years ago

That task here is: A 'compute-slopes' function gets a list of points and returns a list of slops between consecutive points.

The problem: What if points have the same x-value, hence the slope is undefined.

Solution 1 Exceptions: throw new DivisionByZero()

Solution 2 Condition Handling: provide the restarts point to (a) return nil, (b) return 0, and (c) return a user-specified value.

Solution 3 Return Error: return Error("division by zero")

jahewson9• 14 years ago

> why don’t more languages have them?

My guess would be that they are not needed. Exceptions for anything other than program death are something of an anti-pattern, potentially spreading the internal details of some piece of code all the way up the stack.

My choice of implementation would be to pass a discriminator function which takes an error and returns true if the error is recoverable. Then you have the choice of either returning all the errors at the end or reporting each earlier when the discriminator is called.

Edit: That's not to say that exceptions can't be used effectively, they just mostly aren't.

brudgers6• 14 years ago

My understanding of Seibel is that Lisp's condition system is designed to avoid exactly the problem of internal details spreading up the stack.

It does so by allowing conditions (of which exceptions are one type) to be handled by a higher level of the call stack without unwinding the call stack, and allows for the determination of which particular condition handler to be made independently (and at a higher level of the call stack) than the implementation of the handler.

The "anti-pattern" may be a result of the way in which other languages implementation of exception (or condition) handling differs from that of Lisp.

jahewson9• 14 years ago

> [conditions are] handled by a higher level of the call stack without unwinding the call stack, and allow for the determination of which particular condition handler to be made independently (and at a higher level of the call stack) than the implementation of the handler.

Is exactly what the higher-order function I described would let you do. Why do we need conditions?

Edit: the answer to this question is "because we want to use exceptions". But there's really no need to.

stray• 14 years ago

Lisp's condition system allows you to handle the condition without unwinding the stack.

That's the whole point.

jahewson9• 14 years ago

So does a higher-order function...

Therefore it can't be the whole point, because that mechanism already exists, which most commenters here seem to have not realised.

The actual point is, that it gives you more powerful exceptions, but you don't actually need them.

mschaef• 14 years ago

> My choice of implementation would be to pass a discriminator function which takes an error and returns true if the error is recoverable.

Conditions are close to exactly that + syntactic sugar.

`handler-bind` lets you establish a dynamic scope, inside of which specific functions are called when specific conditions are signaled. Signaling a condition = call the first handler function in the list that handles the signaled condition. The handler is called before the stack is unwound, which allows it to elect to either `throw`, invoke a restart, or ignore the condition entirely.

This is in contrast to exceptions in (say) Java. `throw new FooException()` does a lot of potentially destructive work before control gets transfered to the exception handler: It allocates an exception object, which fully captures the current thread's stack. (It's not too common, but this can be bypassed, if you don't throw a new exception object...) It unwinds the stack to the frame that has a matching `catch`.

Only after those two things have happened does the exception handler get a chance to decide how to proceed, and at that point, it's too late for many recovery strategies. Conditions help this by making it easier to intercept the process earlier.

Where they hurt is that they add another level of complexity to API design. With Java style exceptions, if the function fails, it fails completely. There's a convenient simplicity to that. With Lisp style conditions and restarts, A function will let you know how things are going as it does its work and then give you the opportunity to change its behavior midstream. It's a lot more work to design that kind of API, and it's probably well past the point of diminishing returns for most kinds of development.

jahewson9• 14 years ago

Great answer. I'm left wondering if the solution is more complex than the problem.

brudgers6• 14 years ago

Peter Seibel's explanation in the online version of Practical Common Lisp Chapter 19:

http://gigamonkeys.com/book/beyond-exception-handling-condit...

Practical Common Lisp Table of Contents: http://gigamonkeys.com/book/

ajuc2• 14 years ago

Seems like useful feature to have.

Is it supposed to be only used for error handling (like exceptions) or can you use it for unconventional control flow, too?

masklinn10• 14 years ago

You can use it for unconventional control flows if you wish to. For instance, see [0] which implements dynamically scoped variables on top of conditions. I'm not sure that's overly common, but I'm a smalltalk dilettante

[0] http://www.lshift.net/blog/2012/06/27/resumable-exceptions-c...

ken• 14 years ago

Dan Weinreb (one of the Multics users who brought the Multics condition system to Lisp) wrote about why "error" is not a great term: http://danweinreb.org/blog/what-conditions-exceptions-are-re...

Personally, I've not seen it used for arbitrary control flow. If you control the syntax of things inside your block, then macros are probably good enough. If you're passing values outside of your own block, then you probably need continuations.

OTOH, Lisp programmers have a tendency to not care what something is "supposed to be used for". Using Lisp at all is said to be "the most intelligent way to misuse a computer". :-)

ataggart2• 14 years ago

If I understand correctly, the idea is that `parseLog` is a library function, while `parseLogSilently` and `parseLogInteractively` are examples of callers to that library.

If so, then the complexity of documentation and understanding of `parseLog` is pretty high. The caller needs to not only know the different conditions (e.g., `ParseError`), but also the available recovery strategies for those conditions (e.g., `SkipEntry`, `FixEntry`), and further that some of those strategies require an additional caller-provided function (e.g., `askToFixEntry`) with its own separate requirements (e.g., it takes an exception and returns a line).

Why would it not be better, since there are a finite number of provided recovery strategies, to simply hide that complexity within library-provided functions? E.g.:

    (defn parse-log [file] ...) ; aborts on ParseError
    (defn parse-log-skip [file] ...) ; skips bad lines
    (defn parse-log-fix [file fix-fn] ...) ; emits (fix-fn line) for bad lines

lubutu3• 14 years ago

Notice 'parseLogLoudly'. You could not implement that using your set of functions: you are not allowing for other code besides 'resume' inside the 'handle' blocks.

Also, I don't believe the mass duplication of code necessary for your approach is acceptable, even within a library.

ataggart2• 14 years ago

>Notice 'parseLogLoudly'. You could not implement that

Fair point, and that's due to my failure in not providing an optional side-effect function arg to `parse-log-skip`. And this brings up a related point, namely that the author of the lib is still responsible for crafting the API. For example, if the author of `parseLog` had not included `recover FixEntry`, would it be possible for the caller to implement it via a condition system? Would that even be a good idea? Using the pseudo-python, the only approach I can think of would be:

    def parseLogInteractively(file):
      do:
        return parseLog(file)
      handle ParseError, e:
        line = askToFixEntry(e.text)
        retry

That reference to `line` is pretty horrific.

Perhaps this is one of those cases where a simplified example helps convey the concept, but is too simple to show where other approaches would fall short.

>mass duplication of code

I'm not clear on what you mean. Multiple functions with different behaviour seems not to altogether different from one function with implicit switches that change its behaviour.

An Introduction to Condition Handling for Non-Lispers

Related Stories