Wandering Thoughts archives


Link: Parsing: a timeline

Jeffery Kegler's Parsing: a timeline (via) is what it says on the title; it's an (opinionated) timeline of various developments in computer language parsing. There are a number of fascinating parts to it and many bits of history that I hadn't known and I'm glad to have read about. Among other things, this timeline discusses all of the things that aren't actually really solved problems in parsing, which is informative all by itself.

(I've been exposed to various aspects of parsing and it's a long standing interest of mine, but I don't think I've ever seen the history of the field laid out like this. I had no idea that so many things were relatively late developments, or of all of the twists and turns involved in the path to LALR parsers.)

links/ParsingATimeline written at 00:48:04; Add Comment

Go and the pragmatic problems of having a Python-like with statement

In a comment on my entry on finalizers in Go, Aneurin Price asked:

So there's no deterministic way to execute some code when an object goes out of scope? Does Go at least have something like Python's "with" statement? [...]

For those who haven't seen it, the Python with statement is used like this:

with open("output.txt", "w") as fp:
    ... do things with fp ...

# fp is automatically closed by the
# the time we get here.

Python's with gives you reliable and automatic cleanup of fp or whatever resource you're working with inside the with block. Your code doesn't have to know anything or do anything; all of the magic is encapsulated inside with and things that speak its protocol.

Naturally, Go has no equivalent; sure, we have the defer statement but it's not anywhere near the same thing. In my opinion this is the right call for Go, because of two issues you would have if you tried to have something like Python's with in Go.

The obvious issue is that you would need some sort of protocol to handle initialization and cleanup, which would be a first for Go. You need the protocol because a big point of Python's with is that it magically handles everything for you without you having to remember to write any extra code; it's part of the point that using with is easier and shorter than trying to roll your own version (which encourages people to use it). If you're willing to write extra code, Go has everything today in the form of defer().

But beyond that there is a broader philosophical issue that's exposed by Aneurin Price's first question. In a language like Go where your local data may escape into functions you call, what does it mean for something to go out of scope? One answer is that things only go out of scope when there's no remaining reference to them. Unfortunately I believe that this is more or less impossible to implement efficiently without either going to Rust's extremes of ownership tracking in the language or forcing a reference counting garbage collector (where you know immediately when something is no longer referenced). This leaves you with the finalizer problem, where you're not actually cleaning up the resource promptly.

The other answer is that 'going out of scope' simply means 'execution reaches the end of the relevant block'. As in Python, you always invoke the cleanup actions at this point regardless of whether your resource may have escaped into things you've called and thus may still be alive somewhere. This implicit, hidden cleanup is a potentially dangerous trap for your code; if you forget and pass the resource to something that retains a reference to it, you may get explosions (much) later when that now-dead resource is used. If you're in luck, this use is deterministic so you can find it in tests. If you're unlucky, this use only happens in, say, an error path.

Using defer() instead of an implicit cleanup doesn't stop this problem from happening, but it makes explicit what's going on. When you write or see a defer(fp.Close()), you're pointedly reminded that at the end of the function, the resource will be dead. There is no implicit magic, only explicit actions, and hopefully this creates enough warning and awareness. Given Go's design goals, being explicit here as part of the language design makes complete sense to me. You can still get it wrong, but at least the wrongness is more visible.

(I don't think being explicit is necessarily better in general than Python's implicit magic. Go and Python are different languages with different goals; what's appropriate for one is not necessarily appropriate for the other. Python has both language features and cultural features that make with a good thing for it.)

programming/GoVersusPythonWith written at 00:40:28; Add Comment

Page tools: See As Normal.
Login: Password:
Atom Syndication: Recent Pages, Recent Comments.

This dinky wiki is brought to you by the Insane Hackers Guild, Python sub-branch.