2009/12/8 Jon Fairbairn :

Deniz Dogan writes:

...

Has there been any serious suggestion or attempt to change the syntax of Haskell to allow hyphens in identifiers, much like in Lisp languages? E.g. hello-world would be a valid function name.

I (among others) suggested it right at the beginning when we were first defining the lexical syntax, and have raised the possibility again a couple of times now that unicode hyphens are available, but it wasn't liked at the beginning and hasn't excited much interest since.  Why do you want them?

I just like the standard Lisp naming convention better than camel casing and saw someone on #haskell mentioning it the other day. No biggie though. -- Deniz Dogan

2009/12/9 Maciej Piechotka :

On Tue, 2009-12-08 at 10:56 +0100, Deniz Dogan wrote:

...

Has there been any serious suggestion or attempt to change the syntax of Haskell to allow hyphens in identifiers, much like in Lisp languages? E.g. hello-world would be a valid function name.

You mean to parse a - b differently then a-b? You don't have the problem in LISP as AFAIR you use (- a b) but in Haskell it would be a problem.

I understand. How do GHC extensions work? Would a (hopefully tiny) GHC extension make it possible to use normal hyphens (i.e. those in ASCII) in identifiers? If so, is anyone interested in writing one? -- Deniz Dogan

2009/12/9 Richard O'Keefe :

On Dec 9, 2009, at 10:54 PM, Maciej Piechotka wrote:

...

You mean to parse a - b differently then a-b? You don't have the problem in LISP as AFAIR you use (- a b) but in Haskell it would be a problem.

It's a problem that COBOL solved a long time ago:        COMPUTE INCREASED-DEBT = TOTAL-EXPENSES - AFTER-TAX-INCOME. Haskell already has this problem with ".", where we generally need to put spaces around "." with the meaning "composition" and not put spaces around other uses.

This is something someone could easily try out by writing a trivial preprocessor to convert hyphens with letters on each side to underscores.  See how it works.

Given the amazinglyUglyAndUnreadably baStudlyCaps namingStyle that went into Haskell forNoApparentReasonThatIHaveEverHeardOf, it might be nice to have a wee preprocessor that turned        <lower case letter one> _ <lower case letter two> into    <lower case letter one> <Upper case letter two>

so that I could write take_while and Haskell could see takeWhile. [I'm writing this in MacOS X Mail.  "takeWhile" is underlined in red as a spelling mistake, "take_while" is not.  Maybe they know something...]

Here is such a preprocessor.  This is meant for people to try out. I don't claim that it's perfect, it's just a quick hack.

Is there any flag I can pass to e.g. GHC to make it use the preprocessor automagically or do I write my own little hack to apply the preprocessor? -- Deniz Dogan

On Thu, 2009-12-10 at 11:54 +1300, Richard O'Keefe wrote:

On Dec 9, 2009, at 10:54 PM, Maciej Piechotka wrote:

...

You mean to parse a - b differently then a-b? You don't have the problem in LISP as AFAIR you use (- a b) but in Haskell it would be a problem.

It's a problem that COBOL solved a long time ago: COMPUTE INCREASED-DEBT = TOTAL-EXPENSES - AFTER-TAX-INCOME. Haskell already has this problem with ".", where we generally need to put spaces around "." with the meaning "composition" and not put spaces around other uses.

Well. http://en.wikipedia.org/wiki/Backward_compatibility About '.' - there are two contexts in which it is used: - Hierarchical modules: In which first letters (except maybe last one) are capitalized - Composition: In which first letters are not capitalized In combination of those I belive there is no disambiguoty since the lower case words functions as terminators. While it can be discussed as an error of the past it is not the problem that spaces creates problems: ghci> :t head . tail head . tail :: [a] -> a ghci> :t head .tail head .tail :: [a] -> a ghci> :t head.tail head.tail :: [a] -> a Or even: ghci> :t Data.IORef.readIORef.System.IO.Unsafe.unsafePerformIO.Data.IORef.newIORef Data.IORef.readIORef.System.IO.Unsafe.unsafePerformIO.Data.IORef.newIORef :: a -> IO a Even in hugs: Hugs> :t head.tail head . tail :: [a] -> a Hugs> :t head . tail head . tail :: [a] -> a Hugs> :t head .tail head . tail :: [a] -> a Hugs> :t head.tail head . tail :: [a] -> a Main> :t Data.IORef.readIORef.System.IO.Unsafe.unsafePerformIO.Data.IORef.newIORef readIORef . unsafePerformIO . newIORef :: a -> IO a I would question sanity of using . here but it is not the case that we get flood of posts why adding/removing space next to ./- causes any problems. People with practically any programmin backround would expect a-b be equivalent to a - b. Sometimes this 2 spaces are needed to fit in 72/80 columns.

This is something someone could easily try out by writing a trivial preprocessor to convert hyphens with letters on each side to underscores. See how it works.

Given the amazinglyUglyAndUnreadably baStudlyCaps namingStyle that went into Haskell forNoApparentReasonThatIHaveEverHeardOf, it might be nice to have a wee preprocessor that turned <lower case letter one> _ <lower case letter two> into <lower case letter one> <Upper case letter two>

so that I could write take_while and Haskell could see takeWhile. [I'm writing this in MacOS X Mail. "takeWhile" is underlined in red as a spelling mistake, "take_while" is not. Maybe they know something...]

Here is such a preprocessor. This is meant for people to try out. I don't claim that it's perfect, it's just a quick hack.

Personally I don't have any strong feelings about conventions as long as they are consistent within one language. Camel cases are no more uncommon then the underscore and they saved space in the past (ok. now it does not matter) and hyphen is very rarly used (to not have problem with minus). For example: - Java - camel cases both in classes and methods (convention very similar to Haskell) - .Net - the same (but first letter of method is capitalized) - Python - underscore - C/C++ - Veries Regards

On Wed, 2009-12-09 at 15:54 -0800, Gregory Crosswhite wrote:

On Dec 9, 2009, at 3:24 PM, Maciej Piechotka wrote:

...

On Thu, 2009-12-10 at 11:54 +1300, Richard O'Keefe wrote: - Composition: In which first letters are not capitalized

In combination of those I belive there is no disambiguoty since the lower case words functions as terminators.

Except for data constructors, which are a special class of functions that start with capital letters; this where an ambiguity can (and does) crop up that is resolved by the spacing rule.

Cheers, Greg

You're right. However it is still rather rare case comparing to the use of - I belive. Epsecially that with - we already have sufficiently complicated paring rules [($), ($a), (a$) where $ is any operator except - vs. -] Regards

The problem that I see with elision resulting in juxtaposition is that it would alter the layout. Each line would not have the same number of characters before and after the makeover. Visual appearance would also be altered. Replacing hyphens with underscores as I originally proposed would alleviate both of these problems. There is also the matter of canonicalization for the purposes of comparing identifiers. Would it be wise for the a version that lacks hyphens be regarded as equivalent to one that does? -------------------------------------------------- From: "Richard O'Keefe" Sent: 09 Wednesday December 2009 1554 To: "Maciej Piechotka" Cc: Subject: Re: [Haskell-cafe] Re: Allowing hyphens in identifiers

On Dec 9, 2009, at 10:54 PM, Maciej Piechotka wrote:

...

You mean to parse a - b differently then a-b? You don't have the problem in LISP as AFAIR you use (- a b) but in Haskell it would be a problem.

It's a problem that COBOL solved a long time ago: COMPUTE INCREASED-DEBT = TOTAL-EXPENSES - AFTER-TAX-INCOME. Haskell already has this problem with ".", where we generally need to put spaces around "." with the meaning "composition" and not put spaces around other uses.

This is something someone could easily try out by writing a trivial preprocessor to convert hyphens with letters on each side to underscores. See how it works.

Given the amazinglyUglyAndUnreadably baStudlyCaps namingStyle that went into Haskell forNoApparentReasonThatIHaveEverHeardOf, it might be nice to have a wee preprocessor that turned <lower case letter one> _ <lower case letter two> into <lower case letter one> <Upper case letter two>

so that I could write take_while and Haskell could see takeWhile. [I'm writing this in MacOS X Mail. "takeWhile" is underlined in red as a spelling mistake, "take_while" is not. Maybe they know something...]

Here is such a preprocessor. This is meant for people to try out. I don't claim that it's perfect, it's just a quick hack.

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

Interesting point. I'll have to try what you suggested. A proportional font would lessen the need to align things up vertically. It is annoying from a typographical stand point when things are almost aligned, but a tad off and so I must admit that I am guilty as charged. Monospaced fonts encourage this condition. -------------------------------------------------- From: "Evan Laforge" Sent: 09 Wednesday December 2009 1847 To: "John D. Earle" Cc: "Haskell Cafe" Subject: Re: [Haskell-cafe] Re: Allowing hyphens in identifiers

On Wed, Dec 9, 2009 at 3:26 PM, John D. Earle wrote:

...

The problem that I see with elision resulting in juxtaposition is that it would alter the layout. Each line would not have the same number of characters before and after the makeover. Visual appearance would also be altered. Replacing hyphens with underscores as I originally proposed

As long as we're pushing pet peeves: don't line things up vertically then. If you use a newline and add a tab level you can still use layout and not have this problem. You wind up with less indentation too, though occasionally more vertical space. And you can use proportional fonts if you want. And you can use editor indent/dedent commands without the editor having to understand haskell. Etc.

My pet peeve is having to lean on the space bar while squinting at the screen because someone wants to line things up vertically.

Am Mittwoch 09 Dezember 2009 23:54:22 schrieb Richard O'Keefe:

Given the amazinglyUglyAndUnreadably baStudlyCaps namingStyle that went into Haskell forNoApparentReasonThatIHaveEverHeardOf,

mb_t's_bcs the ndrscr_stl is considered far uglier and less readable by others (granted, underscore-style with nonabbreviated words is not unreadable, but still extremely ugly)? 'Tis a matter of personal preference.

On Dec 11, 2009, at 3:05 AM, Miguel Mitrofanov wrote:

Not to mention that in Emacs with glasses-mode enabled camelCase can be made even more readable (my personal favorite is highlighting internal capital letters with bold).

And there is some reason why letters following underscores could not be similarly highlighted? Having to use machine assistance to read baStudlyCaps is pretty damning.

there is enough experimentally determined about reading in general to be certain that visible gaps between words materially improves readability, and internal capital letters harm it.

Here is a (slightly mischievous) proposal. Allow the Unicode non-breaking space character (  in HTML) as a valid character in a varid or conid. :-) Regards, Malcolm

I too thought of using NBSP as a word separator. Unicode is replete with spaces of various widths and usages. I suggest, however, that Haskell code does tend to require careful reading, and that ambiguities (like the word boundaries that baStudlyCapsMakesSoHardToLocateQuickly) that we might tolerate in contexts where the text is fairly predictable are far less tolerable in stuff that requires a close reading. One of the problems with baStudlyCaps is that it can give Stress to the wrong words. Take takeWhile as an example. The "while" bit is humanly recoverable because a function follows it. The thing that you most need to see is "take". But it's the least informative part of the name (While) that is stressed. The "visible blank" U+2423 (Unicode calls it an "Open Box", but its function is to visibly indicate a blank) might␣just␣do, but you might as well use underscores (or hyphens).

x <- take_while some_condition some_list

and

x <- takeWhile someCondition someList

x <- take-while some-condition some-list As someone who is dyslexic, I find both camelCase and dashes far easier to read than underscores. I find it hard to count the words in the underscore version - the underscores and spaces confuse me. Dashes do not, presumably because they are 'within' the line of sight for the letters. However, I can't help feeling that if we use good IDEs, we could 'internationalize' identifiers into english/french/chinese/camel/dash or whatever. Matthew

Am Montag 14 Dezember 2009 01:44:16 schrieb Richard O'Keefe:

On Dec 13, 2009, at 3:44 AM, Daniel Fischer wrote:

...

Am Freitag 11 Dezember 2009 01:20:55 schrieb Richard O'Keefe:

...

On Dec 11, 2009, at 3:00 AM, Daniel Fischer wrote:

...

Am Mittwoch 09 Dezember 2009 23:54:22 schrieb Richard O'Keefe:

...

Given the amazinglyUglyAndUnreadably baStudlyCaps namingStyle that went into Haskell forNoApparentReasonThatIHaveEverHeardOf,

mb_t's_bcs the ndrscr_stl is considered far uglier and less readable by others

Come ON. Argue honestly!

Thanks, but I have to return that compliment.

Not a bit of it. "mb_t's_bcs the ndrscr_stl" was NOT playing fair and you knew it.

1. I wasn't playing in the under_score vs. camelCase game, just proposing a possible reason why the camelCase may have been chosen for Haskell's standard libraries. For that, I deliberately used an unrealistically exaggerated example of unreadable underscores. Even more unrealistic than your camelCase parody. The very next thing, I said that full_word_underscores is *not* unreadable, making clear (apparently not to you, sorry about that) that the example is not to be taken at face value. 2. Below, you say that it was an honest misunderstanding, so it's okay then. I didn't expect that to be the case but suspected intentional misinterpretation.

...

I claim that the part not using insanely abbreviated words (isConsidered, farUglier, byOthers) *is* readable. Also that moderately abbreviated words are readable in camelCase as well as under_score.

I guess we must mean different things by "readable" then.

By readable, I mean (broadly) "easy to read". Of course, some find is_considered far_uglier by_others easier to read, others harder, and yet others find both equally easy to read. I have no difficulty believing that some find either of the styles hard to read, but I don't expect experienced programmers to be in that group.

I've approached my head of department about running an experiment.

...

...

...

(granted, underscore-style with nonabbreviated words is not unreadable, but still extremely ugly)?

Who "grants" that underscore separation with fully written words is "still extremely ugly"? Not me!

Is that remark really unclear,

Yes, it is really unclear. It read as "x (granted that y)".

Aha. English more fragile than thought. Sorry.

For clarity, it should have been "(still extremely ugly, granted that underscore-style with nonabbreviated words is not unreadable)".

No, that is not what I meant. [Granted,] Underscore-style with nonabbreviated words is not unreadable. Even with written-out words, underscore-style is still extremely[1] ugly. [1] This, however, is an exaggeration as far as I'm concerned. I find the underscore-style ugly, but not extremly so. It was your "amazinglyUgly" which caused it to appear in that sentence.

...

Nobody *grants* anything is ugly or not, that is an aesthetic judgment, as such entirely a matter of personal preference - you can agree with it or not.

Where is it written that aesthetic judgements are _entirely_ a matter of personal preference?

I think you could find that written in many texts on aesthetic relativism. Doesn't matter, though. Of course, one's personal preferences aren't developed in a vacuum, they are strongly influenced by genes and society. So a lot of preferences are shared within a culture and even across cultures and aesthetic judgments aren't entirely *individual*. Nevertheless, I'm convinced that there is no Platonic idea Beauty (or Ugliness) and that if A says that van Gogh's Sunflowers is a beautiful picture while B says it's ugly, it's not the case that one is objectively right, the other wrong. Both are judgments based on their respective preferences and nothing else (unless: lying, acting, succumbing to social pressure, other reasons to not express one's actual opinion).

As a student I was in the Archaeological society. One of the things I learned was this: - many ancient cultures would ritually "kill" grave goods - some grave goods would be smashed up, others would just have a chip out of them - as a rule, the ones that were least damaged were the ones the archaeologists considered to be the most beautiful (after repair). If an aesthetic sense about pots can be shared by modern archaeologists and people living five thousand years ago, it's hardly "entirely personal".

Depends on what you mean by "entirely personal". All I'm saying is that one can find a thing beautiful which another finds ugly.

What you may not appreciate is that I have been a Smalltalk programmer for a long time. I've read and written a lot of code like

printSolutions (self sendSolutionsUsing: Empty and: Empty to: [:p :n | Transcript nextPutAll: 'p = '; print: p; nextPutAll: ', n = '; print: n; nextPut: $.; cr. true "stop after reporting first solution"] ) ifFalse: [ Transcript nextPutAll: 'No solutions.'; cr].

If it were just a matter of experience, then this experience should surely have taught me to love baStudlyCaps.

No. It should have tought you to *read* camelCase - unless your aversion is so strong that you actively refuse to learn reading it.

...

...

I have not been able to discover an experimental study of word separation style effect on readability in programming. I've been wondering about running a small one myself, next year. But there is enough experimentally determined about reading in general to be certain that visible gaps between words materially improves readability, and internal capital letters harm it. Now that applies to ordinary text, but until there's evidence to show that it doesn't apply to program sources, it's a reasonable working assumption that it does.

I doubt that.

Until the evidence is in, what other reasonable working assumption is there?

None.

...

Sourcecode is so different from ordinary text (a line of sourcecode rarely contains more than four or five words),

OK, let's try it. Shakespeare sonnet number 1, first four lines, but split into shorter chunks

With spaces:

From fairest creatures we desire increase, That thereby beauty's rose might never die, But as the riper should by time decease, His tender heir might bear his memory:

With underscores:

From fairest_creatures we desire_increase, That thereby beauty's_rose might_never_die, But as the_riper should by_time decease, His tender_heir might_bear his_memory:

With baStudlyCaps:

From fairestCreatures we desireIncrease, That thereby beauty'sRose mightNeverDie, But as theRiper should byTime decease, His tenderHeir mightBear hisMemory:

baStudlyCaps doesn't read any better with short lines.

I have no trouble reading either version. And that although this is not what camelCase is intended for (as far as I know, the purpose of it is to mark word boundaries within *one token* [identifier]).

...

that I'd be very wary to transfer the findings for one to the other.

It's not uncommon for style guides to explicitly recommend that program code should be spaced like text. Apple notoriously violate this in Objective C. Where Smalltalk would have

this sendSolutionsUsing: Empty and: Empty to: that

Objective C practice is to write

[this sendSolutionsUsing:Empty and:Empty to:that]

So? Whitespace helps tokenising and thus increases readability (for me, at least). What's the relation to the question whether camel case and underscore are readable or not?

...

If somebody claimed that of

x <- take_while some_condition some_list

and

x <- takeWhile someCondition someList

either was objectively more readable than the other, I wouldn't believe it without lots and lots of hard evidence.

"Persaude a man against his will, he's of the same opinion still." How _much_ evidence?

Replicated studies with enough participants from enough different environments/cultures showing that more than 99% of the participants find it clearly more readable. That's due to the *objectively*; for such a strong claim, you need unusually strong evidence. If you claim that a [large] majority find one more readable, it's far easier. And if it turns out that a large majority finds underscores more readable, that would be a sufficient reason to use that [for new languages - mixing camel case and underscores in one language I believe to be as bad as usngabbrvtdwrdswocapsorundrscores]. I take the widespread presence of both as an indication that the majority isn't very large, so you'd have a little work to do to convince me.

That's an artificial example. Haskell code doesn't tend to look like that. It's much more likely to be a choice between x' <- takeWhile p x and x' <- take_while p x

And in such code *I* see _no_ difference in readability. Both read flawlessly, without any hiccough. It's when more than two several-word identifiers are juxtaposed that readability degrades (for both). You should've tried enumFromThenToWithCuteExtraForKicks - such identifiers I find actually harder to read in camel case than with underscores. But using such identifiers is evil and wrong anyway.

I'll repeat the run length figures I previously posted today, but this time as percentages

len pct 1 67% 2 22% 3 7% 4 3% 5 1%

...

5 <1%

for me the underscores push the space-separated parts together, so that I have the tendency to bracket it

x <- take (while some) (condition some) list

Such lines are rare. In my sample (very little of which was my own), under 14% of (lines having at least one identifier) had two or more internal capitals.

Let's look at the number of internal capitals per run:

0 99324 79% 1 20761 16% 2 4100 3% 3 791 3..10 all have under 1% each; 4 221 the sum is 2% 5 68 6 11 7 14 8 4 9 0 10 1

takeWhile someCondition someList has three internal capitals, so it's one of just the 5% of lines with three or more. Let's try a more specific question: what proportions of runs have the pattern aB cD eF? 44 out of 125295, or 0.00035. You're asking me to sacrifice readability everywhere else for the sake of one line in every 2850? (Not that I do find that line more readable in basStudlyCaps.)

Not at all. What gave you that idea? Since camel case is the style used in the overwhelming majority of Haskell code and I consider mixing the two styles really bad, I ask you to use that in your public code, even if grudgingly. You prefer to read and write code in underscore style. Others prefer camel case. Without an easy way to convert, at least one group won't be happy. You have already posted a preprocessor to convert between the two styles, I think. That's good. If I can help improving it and making it more usable, I'd be happy to (there are a couple of points where the transformation is not trivial, {-# OPTIONS_GHC #-}, foreign import).

Am Dienstag 15 Dezember 2009 03:04:43 schrieb Richard O'Keefe:

On Dec 14, 2009, at 5:11 PM, Daniel Fischer wrote:

...

1. I wasn't playing in the under_score vs. camelCase game, just proposing a possible reason why the camelCase may have been chosen for Haskell's standard libraries.

But the insanely abbreviated example did not provide such a reason.

Of course not. But if you expand it - and it's not difficult, even when insanely abbreviated -, the resulting sentence gives a possible reason: "Maybe it's because the underscore style is considered far uglier and less readable by others." If the early Haskellers felt that way, isn't it perfectly natural that they chose the camelCase style? Of course, they may have had entirely different reasons, or no concrete reason at all and it just happened.

You still haven't explained what the reason is supposed to be: it can't be that baStudlyCase salvages the readability of abbreviation because it doesn't.

No, nothing salvages insane abbreviation.

Indeed, it makes it worse, because you can't always tell where one abbreviation ends and another begins.

In teaching an information retrieval paper, one of my favourite examples is unionised. Does it mean (union+ise)+ed "having had the workers organised into a union" or un+(ion+ised) "not having had its molecules turned into ions". When I mean the latter, I always write un-ionised.

Yes, some words are ambiguous. Although I too find un-ionised superior, unIonised also works - okay in a serif font, just barely in sans-serif.

Now consider an actual Java class name, where I genuinely didn't know what the answer was. INSURL baStudlyCaps style for Java doesn't allow underscores in class names. (This is actual Sun code.) Is this something to do with insurance? Is this something to do with URLs for the US Immigration and Nationalization Service? Are Inertial Navigation Systems involved? Is the mention of 'URL' anything to do with URLs, or should this be parsed something like (I) (NSU) (RL)? With underscores, the actual parsing, INS_URL, would be unambiguous.

Or take NVList, another real name. Is it an NV_List (where I don't know what NV is), an N_V_List (where I don't know what N and V are), or an N_VList (where I do know what a vlist is). In fact it's a Name_Value_List. I _might_ have had a clue with N_V_List...

Do you agree that both are horrible names regardless of whether one uses camel case or underscores? INSURL even worse than NVList? Although, considering that it's in "package com.sun.corba.se.impl.naming.namingutil" (not that I find that package name particularly well chosen), the URL part is pretty evident. Nevertheless, ugh, and indeed, INS_URL would be better. Still better would be a name at least partially resolving INS (it's not the International Necronautical Society, I hope, that would be beastly to expand). Considering the other, NamedValueList or Named_Value_List would be far better than any sprinkling of underscores over NVList.

My point here is that if you separate words with spaces, dots, hyphens, underscores, backslashes, or almost anything, you are going to have _much_ less trouble with abbreviations than if you just jam them together baStudlyCaps-style.

Agreed, that kind of shitty names are less bad with underscores.

As for my "parody" of baStudlyCaps, thatIsExactlyHowItLooksToMe.

It's not how it looks to me.

...

I think you could find that written in many texts on aesthetic relativism.

They are empirically wrong.

Interesting. I thought the question whether there is an objective quality of beauty was metaphysical and thus it's impossible to be empirically right or wrong in that respect. So, pray tell, what is the physical substrate of beauty, how can one measure it?

...

Both are judgments based on their respective preferences and nothing else

I disagree. Sometimes, people can articulate _why_ they like or dislike things. For example, I like anything spacious and bright. This explains very well why I prefer landscapes (spacious) to portraits (not spacious).

Yes, it does. But it doesn't mean that landscapes are objectively beautiful and portraits are objectively ugly. There are people who prefer portraits to landscapes. They are not wrong, they just have different preferences.

When it comes to depictions of plants, animals, people, and so on, I prefer healthy to unhealthy, friendly appearance to hostile/dangerous.

Yup. Although my favourite is Toulouse-Lautrec and I like Brueghel (P. the elder) and Bosch too, in general I prefer nice and friendly.

Given that, you could probably predict my response to most paintings fairly well. If I and anyone I personally knew disagreed about which of two paintings was "better",

Are we here at the core of a misunderstanding? I'm not talking about "better", only about "beautiful".

I would expect to find that we quickly reached agreement about what features were _present_ to what _degree_ and about the technical standard of the work (on a rather coarse scale, but enough). The differences could be explained by the relative _weights_ we gave to the various features. Just as I have learned how to prepare tea and cook onions so that my wife will enjoy them, although I dislike the one and detest the other, so I would expect to be able to learn how to predict someone's aesthetic taste fairly well.

How can you not love tea?

Maybe we do agree. It wasn't clear whether by "preferences" you meant "weights" or "outcomes". The thing is, if "preferences" means "outcomes", there's no reason to expect that people will ever agree, whereas if it means "weights", then it should be possible to find or construct examples differing in a single feature where two people with different weights will agree on which is better.

By (aesthetic) preferences I mean "what you like", "what you enjoy to look at or listen to". Whether you find forests beautiful or green prairies or the high mountains.

In the same way, when it comes to coding style, it may well be that we are responding to the same objective properties of styles, but weighting them differently. It appears, for example, that we both perceive abbreviation, and we both give it a negative weight.

Not unconditionally, of course. parseURI is a good name, I think you will agree that it's better than parseUniformResourceIdentifier or parse_uniform_resource_identifier. And of course, parse_URI is also better than both in my opinion. enum for enumeration/enumerate - rather good than bad, at least in compositions num, val - why not Abbreviation is a little like goto: handle with care

It is therefore to be expected that given two versions of a program in which the *only* thing changed is the degree and/or nature of abbreviation, we'll agree which is better and which is worse.

Probably.

For me to accept "personal preference" as a final explanation of something would be to accept an end to rational investigation.

Yes, in a way. You can of course investigate how people came to their preferences. When it comes to whether somebody prefers Shostakovich or Mozart, Joy Division or Shakira, Duck or Lobster, you can investigate why, but that's it, in the end, people have their preferences. When they enjoy reading Jane Austen but dislike reading Thomas Mann, it's not because Austen was a good writer and Mann was a bad writer (both were really good), it's just because Austen is more to their taste.

...

...

If it were just a matter of experience, then this experience should surely have taught me to love baStudlyCaps.

No. It should have tought you to *read* camelCase - unless your aversion is so strong that you actively refuse to learn reading it.

Where did you ever get the idea that I can't *read* baStudlyCaps?

Nowhere. I was convinced from the beginning that you can read camel case - with greater ease than you're willing to admit.

Just because I can read it doesn't mean that I can't read something else *better*.

Sure.

Life is hard enough without accepting unnecessary difficulties, even if they are moderately small ones.

...

...

...

Sourcecode is so different from ordinary text (a line of sourcecode rarely contains more than four or five words),

I gave the wrong response to that yesterday. Later in the common room I realised what the perfect answer is:

newspapers are ordinary text, newspaper columns are typically four or five words across.

Then your newspapers have narrower columns than the ones I read. Four or five words happen, but only if at least one of them is a monster like Anlagenmechanikermeister (yes, that's a real word, I found it when counting today). Typically, I counted six to eight, ten or twelve occur too. About 40 characters, more when many i's, t's, l's and f's (and few m's, w's) appear.

The number of words per line is therefore not a useful way to distinguish source code from ordinary text.

How about: Source code is (usually) few words embedded in lots of whitespace. Mostly short lines and short paragraphs. Ordinary text is often a massive block of ink with a little whitespace embedded. Yes, the latter is more a characterisation of newspaper columns, books have more whitespace and are easier on the eye. But books have long lines with many words on them. Source code normally comes in much smaller chunks than ordinary text.

...

...

baStudlyCaps doesn't read any better with short lines.

I have no trouble reading either version. And that although this is not what camelCase is intended for (as far as I know, the purpose of it is to mark word boundaries within *one token* [identifier]).

You missed the point of the example, which was that those words were joined (either by underscores or baStudlyJunctions) which formed sensible units. The junctions were not arbitrary.

Sensible units (although the collation of desire and increase irritates me), but distinct tokens nevertheless. takeWhile is one single token.

[1]

...

So? Whitespace helps tokenising and thus increases readability (for me, at least).

[2]

...

What's the relation to the question whether camel case and underscore are readable or not?

In quotation [1], you concede the argument against baStudlyCaps.

No. We might look at source code in a different way. When I see sendSolutionsUsing, I see one token of the language. When I see this sendSolutionsUsing: Empty and: Empty to: that , I see seven or ten tokens (depends on the staus of : in Smalltalk, it probably should be this sendSolutionsUsing : Empty and : Empty to : that because a small thing like a colon is very effectively hidden by long words directly adjacent - and: and to: would be okay, but I like to be consistent locally). When I see [this sendSolutionsUsing:Empty and:Empty to:that] , I see somebody took a deliberate effort to make tokenising harder.

If white space helps finding units of meaning and thus increases

Whitespace helps separating tokens. Units of meaning comes after that (syntax first, then semantics). They come finer grained (the words making up a compound identifier) and coarser grained (several tokens making up a subexpression, like (sortBy cmp) in sortBy cmp xs) than tokens.

readability for you, then white-or-functionally-white space should help finding units of meaning in program text, and baStudlyCaps should be less readable than separated_words.

The only way to have your cake and eat it is to deny that the words making up a compound identifier _are_ units of meaning that should be perceived as such, or at least the only way that I can see. This seems an odd position to hold.

They are units of meaning, but subordinate to the units of syntax, the tokens. For me, that is, you obviously read code differently.

...

...

"Persaude a man against his will, he's of the same opinion still." How _much_ evidence?

Replicated studies with enough participants from enough different environments/cultures showing that more than 99% of the participants find it clearly more readable.

OK, there is no point in my continuing this. Such a level of study is not practically attainable.

Right. And I've been generous, because strictly speaking, a claim that something is objectively true, goes down the drain with one counterexample. So if every single human, with *one* exception found underscores more readable than camel case, it would not be objectively the case that underscores are more readable.

...

That's due to the *objectively*; for such a strong claim, you need unusually strong evidence.

This is NOT one of those extraordinary claims that require extraordinary evidence.

What? With the word 'objectively' in it, it is an extraordinary claim requiring extraordinary evidence. Replace the word 'objectively' with 'generally', or 'for most', and it becomes an ordinary claim which can be established by ordinary evidence.

It's an entirely humdrum claim that what makes ordinary text more readable makes something strongly resembling ordinary text more readable, and as such, perfectly ordinary experimental evidence should do.

However, I don't think source code strongly resembles ordinary text. Therefore I would like to see some evidence before applying findings about ordinary text to source code. Perfectly ordinary evidence is enough, because what matters is 'generally', not 'objectively'.

...

I take the widespread presence of both as an indication that the majority isn't very large, so you'd have a little work to do to convince me.

You are making the assumption that the word separation style of programmers reflects their OWN initial preference.

No. I am making the assumption that the word separation style of a language reflects the preferences of the language designers and/or the authors of the standard libraries. Further, I don't believe that preference of word separation style is strongly correlated with the capability of designing a programming language or implementing a standard library for a blossoming new language. Thus I tentatively believe that the preferences of language designers do not strongly deviate from the preferences of general programmers. Hence I would expect that a large majority of one preference produces a majority of languages with the preferred style.

I am aware of no reason to believe that. People writing Pascal (which didn't _have_ an underscore because there wasn't one in the 6-bit character set it was designed for) or Smalltalk (which didn't _have_ an underscore because there wasn't one in the 7-bit character set it was designed for) simply didn't have a choice.

Didn't know that. So camel case had a head start.

Java's designers seem to have fairly mindlessly copied Smalltalk, and Java's users _have_ to use

However, if they had a strong preference for underscores, they would probably not hav copied Smalltalk, would they?

Java's vast range of predefined baStudlyCaps identifiers, so in effect have no choice. (Unless like me they have a preprocessor.)

I dare to say that we agree that Java has many advantages over some of its rivals, so that using an uncomfortable word separation style may be compensated for by something else (such as NetBeans or Eclipse, maybe).

Yes, we agree on that. However, for many people, Java's style is not uncomfortable. They may be neutral on the matter, prefer camel case or have a slight preference for underscores. I prefer my braces on the same line as if/while/for, I prefer four space indents (no tabs), but I have no problem with if (condition) { action(); } else { other_action(); }

I'm quite sure that we agree that Haskell has *huge* advantages.

No doubt about that.

The unfortunate word separation style offsets that enough that it's worth programming around (using a preprocessor, for example), but not enough to make me stop using it.

...

...

You're asking me to sacrifice readability everywhere else for the sake of one line in every 2850? (Not that I do find that line more readable in basStudlyCaps.)

Not at all. What gave you that idea?

The form of your argument.

???

...

You prefer to read and write code in underscore style. Others prefer camel case. Without an easy way to convert, at least one group won't be happy.

But there *IS* an easy way to convert.

Yes, all the better.

...

If I can help improving it and making it more usable, I'd be happy to (there are a couple of points where the transformation is not trivial, {-# OPTIONS_GHC #-}, foreign import).

Changes are not made inside {-...-} or "...", only to Haskell identifiers.

I hadn't looked at it before. By the way, I love hspp_nested_comment :: Int -> String -> String -- Int should be safe provided comments are not nested 2147483648 -- levels deep. :D :D :D

There's one bug I'm aware of: --<symbol> is treated as a comment.

Easy to fix. hspp xs@('-':'-':_) = case span (== '-') xs of (dashes,more@(c:_)) | isSymbol c -> case span isSymbol more of (sym,rest) -> dashes ++ sym ++ hspp rest | otherwise -> dashes ++ hspp_eoline_comment more _ -> xs Now, would you be interested in a transformation the other way round, so that you can read other people's code in your preferred style?

Am Mittwoch 16 Dezember 2009 07:56:26 schrieb Colin Paul Adams:

...

...

...

...

...

"Daniel" == Daniel Fischer writes:

Daniel> Now, would you be interested in a transformation the other Daniel> way round, so that you can read other people's code in Daniel> your preferred style?

I would, applied to the output of haddock, at least.

As a pre-alpha version: ------------------------------------------------------------ module Main (main) where import Data.Char (isUpper, isLower, toLower) main :: IO () main = interact unCamel unCamel :: String -> String unCamel ('<':cs) = '<' : inTag cs unCamel (a:bs@(b:cs)) | isLower a && isUpper b = a : '_' : toLower b : unCamel cs | otherwise = a : unCamel bs unCamel cs = cs inTag :: String -> String inTag ('>':cs) = '>' : unCamel cs inTag (a:cs) = a : inTag cs inTag [] = [] ------------------------------------------------------------ Compile, run with ./unCamel < /path/to/HaddockOutput.html > /path/to/Haddock_output.html (or whatever Windows uses to redirect stdin and stdout) This makes a few not-well-founded assumptions about haddock's output. Try it on a couple of files, report bugs. If it works satisfactorily, we can add a front-end to make transformation easy (like, pass an input directory and an output directory on the command line, and it recursively copies the directory content, transforming all .html files, giving you a complete documentation with working links).

On Wed, 2009-12-16 at 14:12 +0000, Colin Paul Adams wrote:

I don't think it's practical to edit all the .cabal packages as they come in to say:

ghc-options: -F -pgmF hspp

and cabal install does not recognize this line if I add to to my ~/.cabal/config file.

Duncan, is there a way this can be done?

I hope not! :-) If you want to view API docs in this alternate style then the right place to insert your hack is in haddock. You can then build all your packages with --haddock-options=--transform_to_underscore_style. Duncan

Am Mittwoch 16 Dezember 2009 15:12:31 schrieb Colin Paul Adams:

I tried it. I'm not all that happy with the resulting uncameling.

For instance,

Database.HaskellDB.Sql.PostgreSQL

goes to

Database.Haskell_dB.Sql.Postgre_sQL

which is uglier than before.

Oy. Didn't think of that. If you can live with Database.Haskell_DB.Sql.Postgre_SQL the fix is easy: unCamel :: String -> String unCamel ('<':cs) = '<' : inTag cs unCamel (a:b:c:cs) | isLower a && isUpper b && isUpper c = a : '_' : b : c : unCamel cs unCamel (a:bs@(b:cs)) | isLower a && isUpper b = a : '_' : toLower b : unCamel cs | otherwise = a : unCamel bs unCamel cs = cs Another point is, what about things like Data.Bits.shiftL would that be preferred as shiftL shift_L shift_l ? Generally, when shall we flatten the hump, 1) always -- not 2) unless followed by an uppercase letter 3) when followed by a lowercase letter I think, 3) is better than 2), things like for_m_ or map_m_ don't look right to me. All of them are easy to implement, the question is what to implement. What about feedO'Houlihan? IMO, it should clearly go to feed_O'Houlihan So, perhaps unCamel :: String -> String unCamel ('<':cs) = '<' : inTag cs unCamel (a:b:c:cs) | isLower a && isUpper b && isLower c = a : '_' : toLower b : c : unCamel cs unCamel (a:bs@(b:cs)) | isLower a && isUpper b = a : '_' : b : unCamel cs | otherwise = a : unCamel bs unCamel cs = cs (gives Database.Haskell_DB.Sql.Postgre_SQL, shift_L, for_M_ [that's not optimal, any ideas what to make of that?], feed_O'Houlihan). Please test, report shortcomings.

I.m not sure how I would write this going the other way, using Richard's hspp pre-processor.

I'd want to write Database.Haskell_DB.Sql.Postgresql, but I'd guess

That would become really complicated. I'm afraid if you want that, you'll have to do it yourself.

I'd have to write it as Database.Haskell_DB.Sql.Postgre_s_q_l or just Database.Haskell_DB.Sql.PostgreSQL :-(

Anyway, I'm having trouble with using Richard hspp. It changes ok_url to okUrl, but in fact the function concerned is named ok_url in Network.URL, so the pre-processor would have to be applied as part of cabal install for all packages.

Oh, great. There we have a preprocessor's nightmare, a package using both styles. Let us think what to do with such things.

I don't think it's practical to edit all the .cabal packages as they come in to say:

ghc-options: -F -pgmF hspp

and cabal install does not recognize this line if I add to to my ~/.cabal/config file.

Duncan, is there a way this can be done?

Am Mittwoch 16 Dezember 2009 16:08:43 schrieb Colin Paul Adams:

Daniel> Database.Haskell_DB.Sql.Postgre_SQL

Actually, I would be semi-happy with Database.Haskell-DB.Sql.Postgre-SQL

(obviously we need an option whether to use hypens or underscores. I prefer hyphens.)

That's no problem. Hyphens would also circumvent the Network.URL problem :) Of course, you'd have to set up a not-too-complicated way to type \x2010 in your editor.

I think I prefer shift-left :-)

Sure, mate, that's absolutely not going to be a bitch :-/ Use (.<<.) = shiftL (.>>.) = shiftR instead?

I'd settle for the last.

Also map_m and fiter_m &c? And map_m_, for_m_ ? Resp. shift-l, map-m, map-m_ ? Would that be 4) flatten, when followed by lowercase letter, nothing or underscore?

Daniel> Generally, when shall we flatten the hump, 1) always -- Daniel> not 2) unless followed by an uppercase letter 3) when Daniel> followed by a lowercase letter

I'm nit sure. i think a more sophisticated rule might be needed.

Okay, let's start to find it. We won't find something which is perfect in all cases, but perhaps something that's good enough in almost all cases and never too ugly. Should aBC... -> a_BC... / a-BC... in all cases? If not, in which cases what other behaviour? Should aBc... -> a_bc... / a-bc... always? If not, what else? aB -> a_b /a-b ? aB' -> a_b' /a-b' ? aB_ -> ? aB'c... -> ? aB'C... -> ?

Daniel> feedO'Houlihan?

Daniel> IMO, it should clearly go to feed_O'Houlihan

Yes, I think so.

Am Mittwoch 16 Dezember 2009 16:45:01 schrieb Ben Millwood:

On Wed, Dec 16, 2009 at 2:55 PM, Daniel Fischer

wrote:

...

unCamel :: String -> String unCamel ('<':cs) = '<' : inTag cs unCamel (a:b:c:cs)    | isLower a && isUpper b && isLower c = a : '_' : toLower b : c : unCamel cs unCamel (a:bs@(b:cs))    | isLower a && isUpper b    = a : '_' : b : unCamel cs    | otherwise                 = a : unCamel bs unCamel cs = cs

Excuse my pedantry, but: writeToList -> write_toList.

Nothing to excuse, thanks for spotting it so quickly.

I think the third equation needs to be:

...

unCamel (a:b:cs@(c:_))

| isLower a && isUpper b && isLower c = a : '_' : toLower b : unCamel cs

so that the third character is not ignored in subsequent parses.

Yep. That comes from posting code that never saw your editor :(

By the way, I like camelCase because I think that in most cases you *don't* want to break identifiers up into their component words - you read and understand what the function does once, and then you use it as a word in its own right. Any resemblance to actual English is really just a mnemonic of sorts.

I'm with you.

On Thu, Dec 17, 2009 at 8:39 PM, Richard O'Keefe wrote:

On Dec 17, 2009, at 4:45 AM, Ben Millwood wrote:

...

By the way, I like camelCase because I think that in most cases you *don't* want to break identifiers up into their component words

My experience has been that in order to make sense of someone else's code you *HAVE* to break identifiers into their component words. With names like (real example) ScatterColorPresetEditor, the eye *can't* take it in at once, and telling the difference between that and ScatterColorPresentEditor would be a pain.  Break them up Ada-style as Scatter_Colour_Preset_Editor and Scatter_Colour_Present_Editor and you're away laughing.

I don't know what experience you could have had that led you to believe that *I* have to break identifiers into component words. I am kind of just being a pedant, but I think that using the word "I" throughout this paragraph instead of "you" will more fairly draw attention to the fact that, throughout it, you are still talking about your own experience. I don't like to get caught up into this kind of argument, but my own opinion is that phrases don't make good words. If we write-text English-language using word-phrase-uninflected as words, no matter-verb the word-spacing, it would be difficulty-hard to read-text. My preferred way to increase the readability of code is to keep names short and *limited in scope*. Luke

Here's another *real* baStudlyCaps identifier taken from a real (and useful) program someone else wrote: WEKA_AttributeSelectionEvaluationRanker_SymmetricalUncert

...

- you read and understand what the function does once, and then you use it as a word in its own right.

I defy anyone to recognise WEKA_AttributeSelectionEvaluationRanker_SymmetricalUncert as a word in its own right.

...

Any resemblance to actual English is really just a mnemonic of sorts.

Take a look at another typical example of the genre: TransformerFactoryConfigurationError "transformer factory configuration error" -- the resemblance to actual English is more than accidental.

A sample of a little over 2000 Java class names had this length distribution.

 1: 10  2: 9  3: 35  4: 55  5: 53  6: 72  7: 97  8: 114  9: 127 10: 158 11: 116 12: 145 13: 150 14: 131 15: 120 16: 109 17: 96 18: 84 19: 56 20: 57 21: 53 22: 53 23: 37 24: 21 25: 29 26: 8 27: 9 28: 5 29: 5 31: 2 33: 3 35: 1 36: 3 37: 2 39: 2 44: 2 47: 1 48: 1 49: 1 50: 1 57: 1

Looking at the number of words in each of these "phrases", the distribution is  1: 334  16.5%  2: 893  44.1%  3: 578  28.6%  4: 161   7.5%  5:  43   2.1%  6:  17   0.8%  7:   8   0.4%

Now on a sample of English text (the WSJ collection), the commonest word length was 3 letters and the longest was 26.  These Java class names are clearly quite unlike the words we're used to reading (which would count against my ideas about readability except that they _are_ quite like English _phrases_).

I don't have similar figures handy for Haskell.  You thought myExamplesLikeThisAreMadeUp?  Nope, things that long and with that many words in the phrase are _not_ rare in seriously baStudly code.

If you can see a 57-character 7-word class name as a single "word", you have very different eyes from me.

_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe

On Thu, Dec 17, 2009 at 10:39 PM, Richard O'Keefe wrote:

On Dec 18, 2009, at 5:00 PM, Luke Palmer wrote:

...

 My preferred way to increase the readability of code is to keep names short and *limited in scope*.

That's good.  Haskell code is chock full of one and two letter highly local identifiers.

I am actually rather ambivalent to this practice. I was referring more to the 4-8 letter, 1-2 word range: functions named "collect" or "diagonals" or "split". A name like this together with a type signature is frequently enough to uniquely identify the function in question; without a good name, not so much. I find one and two letter identifiers the best way in a few situations, eg. in "map (x:xs) = ...", where there is hardly a better name for them ("head" and "tail" perhaps... but it is blindly clear from context what they are...). On the other hand, a function like Martin Escardo's:

find_vii :: (Cantor -> Bool) -> Cantor find_vii p = b where b = id'(\n -> if q n (find_vii(q n)) then Zero else One) q n a = p(\i -> if i < n then b i else if i == n then Zero else a(i-n-1))

is pretty rough reading for me. Luke

On Fri, 18 Dec 2009 16:39:21 +1300, you wrote:

My experience has been that in order to make sense of someone else's code you *HAVE* to break identifiers into their component words. With names like (real example) ScatterColorPresetEditor, the eye *can't* take it in at once, and telling the difference between that and ScatterColorPresentEditor would be a pain. Break them up Ada-style as Scatter_Colour_Preset_Editor and Scatter_Colour_Present_Editor and you're away laughing.

I wouldn't notice the difference between Preset and Present in either case. And in the latter example, my eyes would actually be drawn away from Preset/Present and towards Colour, noticing that it is spelled incorrectly... Count me in the "prefers hyphens" camp, by the way. Steve Schafer Fenestra Technologies Corp. http://www.fenestra.com/

Richard O'Keefe wrote:

(Why do people call baStudly syle "camel"? Is there somewhere in the world a species of camel with three or four humps, like XmlNodeSansChildren?)

camelCase differs from StudlyCaps in the case of the first character (that is, XmlNodeSansChildren is *not* camelCase). The camel name can perhaps be justified by the hump shape when only two or three words are involved. This thread is the only place I've ever heard of baStudlyCaps. -- Live well, ~wren

I used to refer to them as dromedaryCase and BactrianCase, but the names never caught on :(

Daniel Fischer writes:

Am Dienstag 15 Dezember 2009 03:04:43 schrieb Richard O'Keefe:

...

On Dec 14, 2009, at 5:11 PM, Daniel Fischer wrote:

...

1. I wasn't playing in the under_score vs. camelCase game, just proposing a possible reason why the camelCase may have been chosen for Haskell's standard libraries.

But the insanely abbreviated example did not provide such a reason.

Of course not. But if you expand it - and it's not difficult, even when insanely abbreviated -, the resulting sentence gives a possible reason: "Maybe it's because the underscore style is considered far uglier and less readable by others." If the early Haskellers felt that way, isn't it perfectly natural that they chose the camelCase style?

As one of the early Haskellers, I definitely preferred underscores, because my intuition told me that it was closer in appearance to normal English¹ text, and my belief was that even programmers read more English than code. Unfortunately I'm not a very persuasive person, and couldn't argue my case (beyond having underscores /permitted/). The problem is that once people have spent some time using one style or other, their ability to self-analyse their reading of it becomes negligible. "Ugly" and "less readable" become synonymous with "not the style I'm used to", irrespective of actual effects on reading speed. And introspection is a notoriously bad method of anyalysing psychological factors in the first place. This really should have been decided with proper experiments.

Of course, they may have had entirely different reasons, or no concrete reason at all and it just happened.

In the absence of hard data, it only takes a slight bias in exposure among members of the committee to tip a decision the wrong way. [1] and quite a high proportion of other natural languages. -- Jón Fairbairn Jon.Fairbairn@cl.cam.ac.uk

...

...

...

...

"Ketil" == Ketil Malde writes:

>> [1] and quite a high proportion of other natural languages. Ketil> Which makes me wonder - might there be a (natural) language Ketil> bias as well? Sure. Different languages have different orthographical traditions. IMHO a project should adopt the styles of the natural language it uses for comments (assuming that the project leader has mandated a language for comments). The Haskell standard libraries don't do this. There is also the question of whether a programming language should be adaptable to the natural language used in a project (e.g. spelling of keywords). I have tried to advocate that it should, several times in the past. But no-one has ever agreed with me on this. -- Colin Adams Preston Lancashire

On Thu, Dec 10, 2009 at 12:54 AM, Richard O'Keefe wrote:

Given the amazinglyUglyAndUnreadably baStudlyCaps namingStyle that went into Haskell forNoApparentReasonThatIHaveEverHeardOf

Compare: someCoolFunc fstParam sndParam fooBarBazQuux some_cool_func fst_param snd_param foo_bar_baz_quux In the first one, it's easy to see that there are 4 identifiers. But, at least for me, the second one is significantly harder to read; spaces and underscored are too similar. So, while underscores more clearly separate words in a single *identifier* and make it easier to read, in some cases I think they make *code* with many identifiers harder to read for exactly the same reason. In languages where function application looks like "f(x,y,z)" instead of "f x y z" this isn't a problem and I often use underscores. -- Johannes Laire

On Dec 11, 2009, at 11:37 PM, Johannes Laire wrote:

On Thu, Dec 10, 2009 at 12:54 AM, Richard O'Keefe wrote:

...

Given the amazinglyUglyAndUnreadably baStudlyCaps namingStyle that went into Haskell forNoApparentReasonThatIHaveEverHeardOf

Compare:

someCoolFunc fstParam sndParam fooBarBazQuux some_cool_func fst_param snd_param foo_bar_baz_quux

In the first one, it's easy to see that there are 4 identifiers. But, at least for me, the second one is significantly harder to read; spaces and underscored are too similar.

To me the gaps in the second one were clear and distinct. In any case, it's not clear that you *need* to "see that there are 4 identifiers". You need to see the *first* identifier, and then you need to see the second, &c, but let's face it, how often do you need to count the words in a sentence at a glance? And what good does it do to know there are four identifiers if you have to go back with a magnifying class to tell what the identifiers mean? I wrote a little program to search for runs of identifiers without intervening newlines or other tokens. I fed it 56,322 lines of Haskell from various sources. Here's the distribution: Len Tally iCap iCap is the average number of internal 1 84,474 0.21 capital letters in all of the words in 2 27,092 0.32 a run. 3 9,273 0.44 4 3,176 0.45 5 866 0.49 6 165 0.47 7 87 0.36 8 98 0.73 9 42 0.36 10 9 0.22 11 5 0.20 12 7 0.71 13 1 0.00 The longest run was "j then Q i xs ys j else Q j ys xs i" which would be the same under any word separation variant. We see from the "iCap" column that on the whole, the average number of internalCapitals per run is less than one. Looking in some detail at the runs with 4 words, I found that the typical pattern is someHairyGreatFunctionName i x xs with a touch of if okToProceedWith x then In fact, out of 3,176 runs with four words, 2,144 of them didn't have any internal capitals in the first place, so would not be at all affected by a change in word separation style. Having examined the remaining 1032 cases with some care, I honestly couldn't find any cases whose readability was worsened by by underscore separation. Don't forget, some of the arguments are constructors, and using underscore (or hyphen) separation makes it *easier* to see them, because they no longer look like parts of identifiers.