Re: too many lines too long

Lots of good replies in this thread. I just wanted to follow up on my suggestion from earlier and give some insight as to why I suggested this: The _only_ reason I really suggested a hard enforcement is because, roughly speaking, maintaining code is roughly ~infinitely more expensive than actually writing it. Abiding by a column limit, if enforced, seems really really trivial to many people and very stupid to reject a patch over. But it is far more useful to the tons of people who may read it later, because the labor balance is asymmetrical. The author of a patch must spend some amount of effort N writing it, where N is roughly linear in the size of the change. But the 'maintenance effort' for that patch is, instead, at least linear in the _lifetime of the project_. Basically: code is very cheap, but maintenance is extremely costly. All that said, I'm not going to lose sleep over this, and it's not a stylistic hill that's worth dying on. And again, if GHC only had a few occurrences of this, it would be easier to fix, but that's not the case, and hampers contributors a bit. If anything, a more pressing stylistic concern is our odd naming scheme that has very little consistency, and that does make it harder to find and read things I think. :) On Mon, Nov 9, 2015 at 3:15 PM, Austin Seipp wrote:

Something like this might be possible. It'd just require implementing a new arcanist linter, I think, and enabling it in .arclint

In general I really sympathize with this. The problem 90% of people hit is that they touch a line that was *already* over 80 columns, so 'arc lint' warns them and gets annoyed, but they don't want to fix or split up a bunch of stuff to avoid it. It's an issue of having to do 'boring work' which nobody likes, and seems very tedious, regardless of the mechanism of how they do the change.

Really, I'm more inclined to begin a policy of rejecting reviews that do not pass the linter. Exceptions can be made, but in general we need to start *enforcing it* with the red button I think. And it would require us to be more diligent about merging patches quickly to reduce the scope of merge conflicts (because fixing an 80col violation normally, almost always, adds more LOC).

However, there are people who in general think the contribution barrier is already too high, and I fear that enforcing this with a hard rule may make people 'give up' because it seems like a pointless thing to mandate to block their changes. I'm not sure how people feel about that, but it is worth keeping in mind the developer economics.

I hope suggesting the possibility of being more forceful against 80col violations doesn't derail this too much. :)

On Mon, Nov 9, 2015 at 3:02 PM, Richard Eisenberg wrote:

...

Hi devs,

We seem to be uncommitted to the ideal of 80-character lines. Almost every patch on Phab I look through has a bunch of "line too long" lint errors. No one seems to do much about these. And Phab's very very loud indication of a lint error makes reviewing the code harder.

I like the ideal of 80-character lines. I aim for this ideal in my patches, falling short sometimes, of course. But I think the current setting of requiring everyone to "explain" away their overlong lines during `arc diff` and then trying hard to ignore the lint errors during code review is wrong. And it makes us all inured to more serious lint errors.

How about this: after `arc diff` is run, it will count the number of overlong lines before and after the patch. If there are more after, have the last thing `arc diff` outputs be a stern telling-off of the dev, along the lines of

...

Before your patch, 15 of the edited lines were over 80 characters. Now, a whopping 28 of them are. Can't you do better? Please?

Would this be ignored more or followed more? Who knows. But it would sure be less annoying. :)

What do others think?

Thanks, Richard _______________________________________________ ghc-devs mailing list ghc-devs@haskell.org http://mail.haskell.org/cgi-bin/mailman/listinfo/ghc-devs

-- Regards,

Austin Seipp, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/

-- Regards, Austin Seipp, Haskell Consultant Well-Typed LLP, http://www.well-typed.com/