Deprecate `$` as bitwise xor, replace by xor(), as discussed in #18696 #18977

davidavdav · 2016-10-16T22:36:24Z

These two commits implement the replacement of $ (bitwise exclusive or) by xor(). The first commit includes all changes that are enough to pass the Julia unit tests. The second commit implements more changes in Base, found by grepping for ' $ '.

There still are 3851 lines in base with a $, but it is not directly clear if any of them is a xor or not (pun intended). This would require a Julia parser perhaps. It would be rather handy to be able to find all usages of $ as xor in the code.

vchuravy · 2016-10-17T00:21:59Z

I am not a big fan of this change. The shorthand form is really useful for implementing algorithm that require bit-twiddling . I see where the need for deprecation is coming from, but could we atleast then also add a Unicode infix alias? \veebar or \oplus should be the right ones.

davidavdav · 2016-10-17T08:32:27Z

Sure, we can add one of these as a replacement, why not? This PR make searching for xor-usage of $ a little bit easier, although there is no guarantee it is complete.

tkelman · 2016-10-17T16:36:41Z

base/sparse/sparsematrix.jl

@@ -904,7 +904,8 @@ function _ispermutationvalid_permute!{Ti<:Integer,Tp<:Integer}(perm::AbstractVec
 n = length(perm)
 checkspace[1:n] = 0
 for k in perm
- (0 < k <= n) && ((checkspace[k] $= 1) == 1) || return false
+ checkspace[k] = xor(checkspace[k], 1)


this changes the control flow order, doesn't it?

Well spotted. I didn't think about the shortcut behaviour.

carlobaldassi · 2016-10-17T23:36:45Z

could we atleast then also add a Unicode infix alias? \veebar or \oplus should be the right ones.

\veebar seems the correct one. At which point, we might as well have \vee, \wedge etc.

Though it may be worth noting that although having an infix substitute goes a long way, full "feature parity" will only be achieved when #15964 is fixed (cf. this and following).

davidavdav · 2016-10-18T08:15:45Z

It looks like ⊻ already is an operator. (Where is this defined?)

⊻ = xor

should be enough, then?

But anyhow, this PR is not really about the addition of alternative operators. There is some discussion about that in #18696 and references therein. It looks like once there is consensus about the unicode operators, it is easy enough to add them. (And if people feel that should happen in this PR, I am happy to add them.)

Deprecation of $ (with the purpose of freeing it for some syntactic function in the future) needs to happen first, I believe. I've just gone through all $s in base, and at the moment it already appears to have quite varied semantics.

tkelman · 2016-10-18T08:29:29Z

It looks like ⊻ already is an operator. (Where is this defined?)

julia/src/julia-parser.scm

Line 24 in 5722529

 (add-dots '(+ - |\|| ⊕ ⊖ ⊞ ⊟ |++| ∪ ∨ ⊔ ± ∓ ∔ ∸ ≂ ≏ ⊎ ⊻ ⊽ ⋎ ⋓ ⧺ ⧻ ⨈ ⨢ ⨣ ⨤ ⨥ ⨦ ⨧ ⨨ ⨩ ⨪ ⨫ ⨬ ⨭ ⨮ ⨹ ⨺ ⩁ ⩂ ⩅ ⩊ ⩌ ⩏ ⩐ ⩒ ⩔ ⩖ ⩗ ⩛ ⩝ ⩡ ⩢ ⩣)))) 

nalimilan · 2016-10-18T08:35:10Z

Deprecation of $ (with the purpose of freeing it for some syntactic function in the future) needs to happen first, I believe. I've just gone through all $s in base, and at the moment it already appears to have quite varied semantics.

I disagree. It would be much easier to just replace $ with ⊻, and mention this in the deprecation warning, than to remove all uses of $ first in Base and in packages, and only then introduce ⊻.

KristofferC · 2016-10-18T10:48:46Z

You need const if you want to do the definition as above.

simonbyrne · 2016-10-18T11:04:33Z

It might also be worth adding \xor tab completion (I'm never going to remember \veebar)

davidavdav · 2016-10-18T11:27:37Z

It might also be worth adding \xor tab completion (I'm never going to remember \veebar)

That is a good idea.

JeffBezanson · 2016-10-18T19:32:58Z

I've just gone through all $s in base, and at the moment it already appears to have quite varied semantics.

I'm aware of 2; xor (infix) and interpolation (prefix). Are there more?

davidavdav · 2016-10-18T22:45:36Z

I'm aware of 2; xor (infix) and interpolation (prefix). Are there more?

I won't doubt you on Julia parsing or semantics :-)

Just grepping for $ I see a lot of $(...) constructions which probably are just interpolation, there is the odd end-of-string match re, sometimes the operator itself (as in export $)---I was just looking for the infix meaning. I hope to have found them all.

vchuravy

If we are using \veebar we should maybe remove the .\veebar version?

julia/src/julia-parser.scm

Lines 23 to 24 in 5722529

 (define prec-plus (append! '($) 

 (add-dots '(+ - |\|| ⊕ ⊖ ⊞ ⊟ |++| ∪ ∨ ⊔ ± ∓ ∔ ∸ ≂ ≏ ⊎ ⊻ ⊽ ⋎ ⋓ ⧺ ⧻ ⨈ ⨢ ⨣ ⨤ ⨥ ⨦ ⨧ ⨨ ⨩ ⨪ ⨫ ⨬ ⨭ ⨮ ⨹ ⨺ ⩁ ⩂ ⩅ ⩊ ⩌ ⩏ ⩐ ⩒ ⩔ ⩖ ⩗ ⩛ ⩝ ⩡ ⩢ ⩣))))

vchuravy · 2016-10-19T08:09:58Z

base/associative.jl

@@ -252,7 +252,7 @@ const hasha_seed = UInt === UInt64 ? 0x6d35bb51952d5539 : 0x952d5539
 function hash(a::Associative, h::UInt)
 h = hash(hasha_seed, h)
 for (k,v) in a
- h $= hash(k, hash(v))
+ h = h ⊻ hash(k, hash(v))


You could add ⊻= in

julia/src/julia-parser.scm

Line 11 in 5722529

'(:= => ~ $=)))

OK, sure, thanks, but is this something for #15964, or is that about a more general approach?

Should I also deprecate $= there then, to really free $ for future syntactical purposes?

simonbyrne · 2016-10-19T08:13:06Z

I don't think we should include ∧ and ∨ in this PR: see #6582 (comment) and #17472 for some discussion of this issue.

simonbyrne · 2016-10-19T08:17:33Z

Also, I think we should keep \veebar, there's no problem with having 2 aliases

davidavdav · 2016-10-24T11:33:08Z

All right, I rebased everything a few times---things should be in line with master-as-we-write. I kept ~~three~~ two commits, so there is a choice what to include (either one or two ~~or all three~~ commits)

63bbaef, deprecates $ and replaces infix-$ occurrences in base with xor()
e296b3e, introduces ⊻ and ⊻=, and replaces most of the above xor()s in base with these. The result of these two commits compares most favourably with master, as it includes mainly $⟶⊻ diffs.

e4a340c, also introduces ∨ and ∧ as synonyms to | and &, as some people in this discussion suggested. This change is still under debate, I gather, so this commit can easily be cherry-unpicked.

I hope this rebase satisfies everyone who have taken part in this discussion.

tkelman · 2016-10-24T18:25:39Z

please separate the third commit to a different pull request

davidavdav · 2016-10-24T23:03:36Z

please separate the third commit to a different pull request

No problem, I removed the third commit from this PR and updated the comment above accordingly.

tkelman · 2016-10-26T20:45:33Z

base/show.jl

@@ -449,7 +449,7 @@ const uni_ops = Set{Symbol}([:(+), :(-), :(!), :(¬), :(~), :(<:), :(>:), :(√)
 const expr_infix_wide = Set{Symbol}([
 :(=), :(+=), :(-=), :(*=), :(/=), :(\=), :(^=), :(&=), :(|=), :(÷=), :(%=), :(>>>=), :(>>=), :(<<=),
 :(.=), :(.+=), :(.-=), :(.*=), :(./=), :(.\=), :(.^=), :(.&=), :(.|=), :(.÷=), :(.%=), :(.>>>=), :(.>>=), :(.<<=),
- :(&&), :(||), :(<:), :(=>), :($=)])
+ :(&&), :(||), :(<:), :(=>), :($=), :(⊻=)])


guess $= should be marked somehow as "remove from this list when deprecation is removed," right?

Yes, you are right there. I left $= in there because it may still be used and it should be shown correctly in expressions.

Can you add a comment as Tony mentioned? Something like

# $= was previously used for xor, issue #18977

tkelman · 2016-10-26T20:47:08Z

src/julia-parser.scm

@@ -8,7 +8,7 @@
 ;; be an operator.
 (define prec-assignment
 (append! (add-dots '(= += -= *= /= //= |\\=| ^= ÷= %= <<= >>= >>>= |\|=| &=))
- '(:= => ~ $=)))
+ '(:= => ~ $= ⊻=)))


⊻= can be under add-dots now right? I think $= wasn't added there mostly because $ has other meanings that were making it complicated to parse correctly

I am not sure---I have to admit that my lisp is not up to scratch. From the code I understand that the add-dots operators add a dotted version as well. ⊻= already works element wise. But there is no broadcasting .⊻ indeed. Should that be added?

davidavdav · 2016-10-30T09:44:23Z

src/julia-parser.scm

@@ -78,7 +78,7 @@
 ; operators that are special forms, not function names
 (define syntactic-operators
 (append! (add-dots '(= += -= *= /= //= |\\=| ^= ÷= %= <<= >>= >>>= |\|=| &=))
- '(:= --> $= => && |\|\|| |.| ... ->)))
+ '(:= --> $= ⊻= => && |\|\|| |.| ... ->)))


@tkelman I suppose this ⊻= should then also move one line up to add-dots, right?

davidavdav · 2016-10-30T09:45:33Z

src/julia-syntax.scm

@@ -2060,6 +2060,7 @@
 '&= lower-update-op
 '.&= lower-update-op
 '$= lower-update-op
+ '⊻= lower-update-op


@tkelman ...and here I should then add a line with '.⊻= ...

davidavdav · 2016-10-30T14:43:57Z

Weekly rebase, to keep up with master.

5e853bb deprecates $ and replaces infix-$ occurrences in base with xor()
ce94dff introduces ⊻ and ⊻=, and replaces most of the above xor()s in base with these. The result of these two commits compares most favourably with master, as it includes mainly $⟶⊻ diffs.

This rebase includes the latest changes suggested in a review by @tkelman.

Incidentally, I am planning another PR that will introduce .&, .| and .⊻ broadcasting operators, which will automatically enable the use of .&=, .|= and .⊻=.

xor.() already automagically works in this commit thanks to #15032.

davidavdav · 2016-11-08T20:50:00Z

Weekly rebase, to keep up with master.

simonbyrne · 2016-11-10T11:01:27Z

Can you add an item to NEWS.md?

simonbyrne · 2016-11-10T18:40:48Z

Can you resolve the conflict with NEWS.md? Then I think this is ready to go.

This includes: - occurrances of $ (as infix operator) and $= in base - occurances of $ in test - the definition in the manual - the entry in REPL help This commit does not introduce an alternative, that will happen in the next commit.

Please note that `⊻=` is in the add-dots group in the parser, preparing for `.⊻=`. However, this commit does not implement `.⊻` yet, this should happen in a separate commit with `.&` and `.|` together.

davidavdav · 2016-11-11T23:27:37Z

The build failure on linux is not related to this PR, I think.

musm · 2016-11-14T04:17:07Z

although I'm against this change, it seems strange to just deprecate $, while still keeping & and |. If this change passes, It would make sense to change all three for consistency, otherwise it's confusing that & behaves as you expect, but $ doesn't.

simonbyrne · 2016-11-14T11:33:44Z

Done! Thanks for your patience, modifying that many files in one go is always a bit tricky.

simonbyrne · 2016-11-14T11:47:14Z

@musm The distinction is that & and | have a long history as bitwise and/or (dating back to C at least). The use of $ for xor on the other hand was very much nonstandard (C uses ^, which we already use for raising to a power), and so I don't think changing this affect anyone's expectations.

JuliaLang/julia#18977

musm · 2016-11-14T16:16:54Z

fair point. Doesn't someone have to update the syntax highlighting on github etc. for this change? It used to be that $ got highlighted.

mgkuhn · 2017-01-27T11:11:15Z

In the computing literature, I encounter ⊕ (U+2295 CIRCLED PLUS) far more commonly as the XOR operator [same as addition over GF(2^n)] than ⊻ (U+22BB). How about supporting both?

nalimilan · 2017-01-27T11:14:30Z

In general it's not a good idea to support multiple ways of doing the same thing in a language (an exception to this is ASCII vs. Unicode operators for practical reasons, but at least this is a consistent rule). You can define it for yourself if you really find ⊕ better.

tkelman · 2017-01-27T13:30:55Z

U+22BB is called xor in unicode. I see ⊕ used for dilation/Minkowski sum pretty often, it's context-dependent and a useful operator to leave available since it doesn't have a single unambiguous interpretation.

carlobaldassi · 2017-01-27T13:58:02Z

Besides what @nalimilan and @tkelman wrote, I would guess that the reason why ⊕ is commonly used for xor in the literature is that for 0/1 variables xor is the same as "addition modulo 2". However, when xor'ing 2 integers you expect a bitwise xor, so ⊕ doesn't seem right to me in that context.

mgkuhn · 2017-01-27T18:08:40Z

@tkelman I would only suggest to define ⊕ as carry-less addition (aka XOR) for Unsigned or Integer, as the operator makes no sense for other types (e.g. floating point). That excludes conflicts with non-integer uses of the same symbol, e.g. the ones in relativity you mention.

@carlobaldassi If you cut off the carry-over bits in n-bit modular integer arithmetic, the resulting operations are called binary Galois Field arithmetic, often abbreviated as GF(2^n). The GF(2^n) addition is exactly the n-bit XOR over unsigned integers. Modulo 2 arithmetic is just a special case n=1 of binary Galois Field arithmetic, commonly labeled GF(2).

Galois Field arithmetic is widely used in cryptography and in error-correction algorithms, as it has similar algebraic properties as integers, without the hardware cost of having to wait for carry-over ripple in your adder. Intel and AMD CPUs since 2010 have included a carry-less multiplication instruction for it.

I would suggest to define in Julia the operators ⊕ and ⊗ for carry-less addition (aka XOR) and carry-less multiplication (CLMUL) between unsigned integers.

carlobaldassi · 2017-01-27T18:28:11Z

@mgkuhn that was exactly my point: addition on GF(n) and xor are only equivalent for n=2. When you do e.g. xor(a::UInt, b::UInt) you expect bitwise xor instead, treating the UInts as bit masks. So ⊕ is not suited for that. For operations on GF(n), you'd better implement your own integer type with n as a type parameter, at which point you can just overload + and *.

StefanKarpinski · 2017-01-27T20:50:22Z

There are also many other possible meanings of ⊕ and ⊗ in other areas (e.g. ⊗ as a tensor product). You're free to define these operators in your own projects, but it's inappropriate for the base language to force specific meanings on symbols that have multiple standard meanings.

JuliaLang/julia#18977

…Lang#18696 (JuliaLang#18977) * Deprecate $ for xor() (JuliaLang#18696) This includes: - occurrances of $ (as infix operator) and $= in base - occurances of $ in test - the definition in the manual - the entry in REPL help This commit does not introduce an alternative, that will happen in the next commit. * Add ⊻ as alternative to xor, and `⊻=` as alternative to `$=`. Please note that `⊻=` is in the add-dots group in the parser, preparing for `.⊻=`. However, this commit does not implement `.⊻` yet, this should happen in a separate commit with `.&` and `.|` together. * Add deprecation comment in show.jl and deprecation item in NEWS.md

tkelman reviewed Oct 17, 2016

View reviewed changes

vchuravy reviewed Oct 19, 2016

View reviewed changes

davidavdav force-pushed the deprecate-dollar branch from 0b31ae0 to e4a340c Compare October 24, 2016 11:15

davidavdav force-pushed the deprecate-dollar branch from e4a340c to e296b3e Compare October 24, 2016 22:47

tkelman reviewed Oct 26, 2016

View reviewed changes

davidavdav commented Oct 30, 2016

View reviewed changes

davidavdav force-pushed the deprecate-dollar branch from e296b3e to ce94dff Compare October 30, 2016 11:56

carlobaldassi mentioned this pull request Nov 7, 2016

highlighting of $(...) JuliaEditorSupport/julia-vim#83

Closed

davidavdav force-pushed the deprecate-dollar branch from ce94dff to 91e01b1 Compare November 8, 2016 20:48

simonbyrne added this to the 0.6.0 milestone Nov 10, 2016

davidavdav added 3 commits November 11, 2016 12:31

Deprecate $ for xor() (JuliaLang#18696)

a9a3076

This includes: - occurrances of $ (as infix operator) and $= in base - occurances of $ in test - the definition in the manual - the entry in REPL help This commit does not introduce an alternative, that will happen in the next commit.

Add ⊻ as alternative to xor, and ⊻= as alternative to $=.

8444b3a

Please note that `⊻=` is in the add-dots group in the parser, preparing for `.⊻=`. However, this commit does not implement `.⊻` yet, this should happen in a separate commit with `.&` and `.|` together.

Add deprecation comment in show.jl and deprecation item in NEWS.md

e24df22

davidavdav force-pushed the deprecate-dollar branch from 499ab71 to e24df22 Compare November 11, 2016 11:51

davidavdav mentioned this pull request Nov 12, 2016

& does not broadcast, and there is no .& (similar problem for "or") #6915

Closed

simonbyrne merged commit 2751493 into JuliaLang:master Nov 14, 2016

simonbyrne added a commit to JuliaLang/Compat.jl that referenced this pull request Nov 14, 2016

Deprecation of $ in favour of xor

1e3aca9

JuliaLang/julia#18977

simonbyrne mentioned this pull request Nov 14, 2016

Deprecation of $ in favour of xor JuliaLang/Compat.jl#289

Merged

simonbyrne added a commit to JuliaLang/Compat.jl that referenced this pull request Nov 14, 2016

Deprecation of $ in favour of xor

7b2a9e2

JuliaLang/julia#18977

davidavdav deleted the deprecate-dollar branch November 19, 2016 13:47

davidavdav restored the deprecate-dollar branch January 27, 2017 15:29

davidavdav deleted the deprecate-dollar branch January 27, 2017 15:35

dpsanders pushed a commit to dpsanders/Compat.jl that referenced this pull request Feb 1, 2017

Deprecation of $ in favour of xor

90d5e46

JuliaLang/julia#18977

stevengj mentioned this pull request May 3, 2017

RFC: Make and & or aliases for && and ||. #19788

Closed

2 tasks

davidavdav mentioned this pull request Jan 26, 2018

Support dict.field syntax for Dict{Symbol, <:Any} #25750

Open

	(define prec-plus (append! '($)
	(add-dots '(+ - \|\\|\| ⊕ ⊖ ⊞ ⊟ \|++\| ∪ ∨ ⊔ ± ∓ ∔ ∸ ≂ ≏ ⊎ ⊻ ⊽ ⋎ ⋓ ⧺ ⧻ ⨈ ⨢ ⨣ ⨤ ⨥ ⨦ ⨧ ⨨ ⨩ ⨪ ⨫ ⨬ ⨭ ⨮ ⨹ ⨺ ⩁ ⩂ ⩅ ⩊ ⩌ ⩏ ⩐ ⩒ ⩔ ⩖ ⩗ ⩛ ⩝ ⩡ ⩢ ⩣))))

Deprecate $ as bitwise xor, replace by xor(), as discussed in #18696 #18977

Deprecate $ as bitwise xor, replace by xor(), as discussed in #18696 #18977

Conversation

davidavdav commented Oct 16, 2016

vchuravy commented Oct 17, 2016

davidavdav commented Oct 17, 2016

tkelman Oct 17, 2016 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

carlobaldassi commented Oct 17, 2016

davidavdav commented Oct 18, 2016

tkelman commented Oct 18, 2016

nalimilan commented Oct 18, 2016

KristofferC commented Oct 18, 2016

simonbyrne commented Oct 18, 2016

davidavdav commented Oct 18, 2016

JeffBezanson commented Oct 18, 2016

davidavdav commented Oct 18, 2016

vchuravy left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

simonbyrne commented Oct 19, 2016 • edited Loading

simonbyrne commented Oct 19, 2016

davidavdav commented Oct 24, 2016 • edited Loading

tkelman commented Oct 24, 2016

davidavdav commented Oct 24, 2016

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

davidavdav commented Oct 30, 2016

davidavdav commented Nov 8, 2016

simonbyrne commented Nov 10, 2016

simonbyrne commented Nov 10, 2016

davidavdav commented Nov 11, 2016

musm commented Nov 14, 2016

simonbyrne commented Nov 14, 2016

simonbyrne commented Nov 14, 2016

musm commented Nov 14, 2016

mgkuhn commented Jan 27, 2017

nalimilan commented Jan 27, 2017

tkelman commented Jan 27, 2017

carlobaldassi commented Jan 27, 2017

mgkuhn commented Jan 27, 2017 • edited Loading

carlobaldassi commented Jan 27, 2017

StefanKarpinski commented Jan 27, 2017

Deprecate `$` as bitwise xor, replace by xor(), as discussed in #18696 #18977

Deprecate `$` as bitwise xor, replace by xor(), as discussed in #18696 #18977

tkelman Oct 17, 2016 •

edited

Loading

simonbyrne commented Oct 19, 2016 •

edited

Loading

davidavdav commented Oct 24, 2016 •

edited

Loading

mgkuhn commented Jan 27, 2017 •

edited

Loading