Automatization for fetching inputs, use nightly, logic simplifications #1
Conversation
Signed-off-by: Manuel Saelices <[email protected]>
Signed-off-by: Manuel Saelices <[email protected]>
Signed-off-by: Manuel Saelices <[email protected]>
Signed-off-by: Manuel Saelices <[email protected]>
This needs for this PR to be merged into the nightly release, as the negative steps are not working in Mojo: modularml/mojo#3650 Signed-off-by: Manuel Saelices <[email protected]>
It's slower but more readable. It will be faster overtime as Mojo dicts and strings become fasters. E.g. see modularml/mojo#3436 modularml/mojo#3528 modularml/mojo#3615 Signed-off-by: Manuel Saelices <[email protected]>
Signed-off-by: Manuel Saelices <[email protected]>
Signed-off-by: Manuel Saelices <[email protected]>
Signed-off-by: Manuel Saelices <[email protected]>
Signed-off-by: Manuel Saelices <[email protected]>
The unsafe pointers have relaxed signatures in the .store() method now
There was a problem hiding this comment.
First of all, thanks for helping with cleaning some of this stuff !
Not sure how you got here but I appreciate it.
Most of this change is very simple - the only part I have some reservations about is the rewrite of day2 to utilize builtin parsing functions, which are still really slow, PTAL at the comment below.
Other than that, should need just some minor fixes.
| for i in range(len(words)): | ||
| self.add(words[i]) | ||
| for w in words: | ||
| self.add(w[]) |
There was a problem hiding this comment.
Thanks - I fixed the language problems when upgrading to 24.5 but never got around to removing these relics.
| @@ -89,11 +84,11 @@ fn main() raises: | |||
| a1 += lsum | |||
|
|
|||
| # Construct matchers for all words. When looking backwards, the words have to be reversed. | |||
| # Fun fact - VariadicList apparently can hold literals, but cannot hold Strings. | |||
| # Fun fact - List apparently can hold literals, but cannot hold Strings. | |||
There was a problem hiding this comment.
I think this is no longer true (used to be way back when) -> please remove comment if migrating to List<> works now
| cnt = int(toks[i * 2]) | ||
| col = toks[i * 2 + 1] | ||
| counter[col] = max(counter[col], cnt) | ||
| return counter |
There was a problem hiding this comment.
While I appreciate the direction, this repo's point is to build fast solutions. While it's nice that Mojo now has some collection builtins, split() was there before and I'm not using it since it's abysmally slow.
This task is a perfect example of that problem - with this approach it's about 8 times slower than the 'hacky but fast' solution was (which was on par with PyPy), and it's now ~2.5 times slower than Python.
The hand-crafted parsers that work over raw memory are not really easy (or efficient) in Python, so this is still some way in which I was able to squeeze some performance of Mojo. OTOH sure, this approach shows performance gaps in the std library. Perhaps add this code as a _slow variant of this solution to highlight this, but keep this version as-is (or modified to adjust for some QoL improvements in recent Mojo as in other bits.
| try: | ||
| c = ord(lines[y][x]) | ||
| except IndexError: | ||
| c = 0 |
There was a problem hiding this comment.
So this is probably more of an effect of assumptions on what is contained in lines and how they are loaded - the code in Python is rather simple and does not do the trailing newline check. Rather than a try/catch here you could fix this by removing the last element of lines[] after it's loaded, if it's empty.
| @@ -1,6 +1,6 @@ | |||
| [project] | |||
| authors = ["Staszek Pasko <[email protected]>"] | |||
| channels = ["conda-forge", "https://conda.modular.com/max"] | |||
| channels = ["conda-forge", "https://conda.modular.com/max-nightly"] | |||
There was a problem hiding this comment.
My experiences with nightly have been mixed recently and would prefer this repo to work under the stable API... that said 24.5 has not been very stable as well....
Are there any benefits to nightly I don't know about ?
| return Span[UInt8, __lifetime_of(self)]( | ||
| unsafe_ptr=self.ptr, | ||
| len=self.size, | ||
| ) |
There was a problem hiding this comment.
An unrelated question but this code suffered a bit from lifetime-related issues... are these lifetime parametrizations useful enough now to tell Mojo that eg. keep stuff around in memory more consistently ?
I understand here it's just needed because Spans work like that, but I never figured out how to make the StringSlices here behave in such a way to make sure their lifetime prolonged the lifetime of the parser they originated from (since they depend on that memory)
day01.mojolooping by usingSpan. This needs for this PR to be merged into the nightly release, as the negative steps are not working in Mojo: [stdlib] Span slicing with negative step modularml/mojo#3650day02.mojo. It's slower but more readable. It will be faster overtime as Mojo dicts and strings become fasters.day03.pystore()in thearray.mojofile The unsafe pointers have relaxed signatures in the.store()method now