Hey there.
Is there new homework on the horizon?
I personally would love it if there was at least one bit of homework per week available. Some people might like to "code along" but to me that feels a little too open ended to get myself to do it on the regular. I like the focused homework assignments because I feel they deliver more learning/hour even if there is (in absolute terms) more to learn by coding along. The latter feels like too much of a time investment for me.
Of course this is just a suggestion, I very much enjoy this project either way :-)
Edit:
I was just having a quick think about this. The problem with following along for me is that weeks, months or years down the road there will likely be large stretches in which I don't have the time/energy/whatever to follow the project. Afterwards I would be sitting there with a codebase that's far behind Per's and it would take quite a lot just to catch up. I think I'm not the only one for whom this would be a certainty were they to follow along.
Now, my idea is:
Maybe you could from time to time (or however often you like, of course) create the homework in a way in which you take the bitwise codebase but strip out whatever is the object of the exercise. Be it an algorithm, a specialized datastructure, a whole subsystem etc. People who are following along could still do the exercise in their own codebase.
Is there new homework on the horizon?
I personally would love it if there was at least one bit of homework per week available. Some people might like to "code along" but to me that feels a little too open ended to get myself to do it on the regular. I like the focused homework assignments because I feel they deliver more learning/hour even if there is (in absolute terms) more to learn by coding along. The latter feels like too much of a time investment for me.
Of course this is just a suggestion, I very much enjoy this project either way :-)
Edit:
I was just having a quick think about this. The problem with following along for me is that weeks, months or years down the road there will likely be large stretches in which I don't have the time/energy/whatever to follow the project. Afterwards I would be sitting there with a codebase that's far behind Per's and it would take quite a lot just to catch up. I think I'm not the only one for whom this would be a certainty were they to follow along.
Now, my idea is:
Maybe you could from time to time (or however often you like, of course) create the homework in a way in which you take the bitwise codebase but strip out whatever is the object of the exercise. Be it an algorithm, a specialized datastructure, a whole subsystem etc. People who are following along could still do the exercise in their own codebase.
Edited by mathk
on
Thanks for the reminder. I'll start thinking about some new assignments.
Finished defers today (I didn't forbid assignments, only declarations, breaks and continue).
Continuing with the idea of assignments(from `if`s and `while`s),
I've added "initializer" to defer so it's possible to capture values manually:
Compare to
where program would double free the same pointer
Idea is to have automatic captures someday and it's a scaffold for it which can be used manually for now. (On C side it just creates variable name, based on which p is used).
From other things. I've added a new modifier - @essential. A function is essential, if its return value can't be ignored. It's like __attribute__((warn_unused_result)) in gcc, only not a warning, but an error.
And asserts now work more or less and can print ast(that was used before c codegen) and values of expressions if something goes wrong or you just ask it to print everything. They support only integers now though, but still now I can test compiler using it itself
which I'll probably do more in a future. Testing from inside of C is pain and suffering.
Continuing with the idea of assignments(from `if`s and `while`s),
I've added "initializer" to defer so it's possible to capture values manually:
1 2 | mem := malloc(10); defer(p:=mem) {free(p);} mem = malloc(100); defer(p:=mem) {free(p);} |
Compare to
1 2 | mem := malloc(10); defer {free(mem);}
mem = malloc(100); defer {free(mem);}
|
where program would double free the same pointer
Idea is to have automatic captures someday and it's a scaffold for it which can be used manually for now. (On C side it just creates variable name, based on which p is used).
From other things. I've added a new modifier - @essential. A function is essential, if its return value can't be ignored. It's like __attribute__((warn_unused_result)) in gcc, only not a warning, but an error.
And asserts now work more or less and can print ast(that was used before c codegen) and values of expressions if something goes wrong or you just ask it to print everything. They support only integers now though, but still now I can test compiler using it itself
which I'll probably do more in a future. Testing from inside of C is pain and suffering.
Nice! I like the capture idea.
It's cool to see all the work you're doing on your parallel implementation.
I'll add asserts today since you reminded me. I was planning on piggybacking on # notes since those take expression arguments syntactically. So #assert(x >= 0) and so on. Right now I only support # notes as top-level decls but the plan is to add them at statement level as well. Regarding the printing, if AST nodes have full start/end string pointer ranges into the stream, you can do a verbatim string copy. I'm only tracking the start right now, but this is a good excuse to fix that.
It's cool to see all the work you're doing on your parallel implementation.
I'll add asserts today since you reminded me. I was planning on piggybacking on # notes since those take expression arguments syntactically. So #assert(x >= 0) and so on. Right now I only support # notes as top-level decls but the plan is to add them at statement level as well. Regarding the printing, if AST nodes have full start/end string pointer ranges into the stream, you can do a verbatim string copy. I'm only tracking the start right now, but this is a good excuse to fix that.
Edited by Per Vognsen
on