Squeak-dev November 2014

squeak-dev@lists.squeakfoundation.org

41 participants
125 discussions

Float hierarchy for 64-bit Spur
by Eliot Miranda 24 Nov '14

24 Nov '14

Hi All, 64-bit Spur can usefully provide an immediate float, a 61-bit subset of the ieee double precision float. The scheme steals bits from the mantissa to use for the immediate's 3-bit tag pattern. So values have the same precision as ieee doubles, but can only represent the subset with exponents between 10^-38 and 10^38, the single-precision range. The issue here is how to organize the class hierarchy. The approach that looks best to me is to modify class Float to be an abstract class, and add two subclasses, BoxedFloat and SmallFloat, such that existing boxed instances of Float outside the SmallFloat range will become instances of BoxedFloat and instances within that range will be replaced by references to the relevant SmallFloat. With this approach ... - Float pi etc can still be used, even though they will answer instances of SmallFloat. But tests such as "self assert: result class == Float." will need to be rewritten to e.g. "self assert: result isFloat". - BoxedFloat and SmallFloat will not be mentioned much at all since floats print themselves literally, and so the fact that the classes have changed won't be obvious. - the boxed Float primitives (receiver is a boxed float) live in BoxedFloat and the immediate ones live in SmallFloat. Making SmallFloat a subclass of Float poses problems for all the primitives that do a super send to retry, since the boxed Float prims will be above the unboxed ones and so the boxed ones would have to test for an immediate receiver. An alternative, that VW took (because it has both Float and Double) is to add a superclass, e.g. LimitedPrecisionReal, move most of the methods into it, and keep Float as Float, and add SmallFloat as a subclass of LimitedPrecisionReal. Then while class-side methods such as pi would likely be implemented in LimitedPrecisionReal class, sends to Float to access them find them via inheritance. An automatic reorganization which moves only primitives out of LimitedPrecisionReal is easy to write. Thoughts? -- best, Eliot

7 26

The Trunk: Monticello-bf.604.mcz
by commits＠source.squeak.org 24 Nov '14

24 Nov '14

Bert Freudenberg uploaded a new version of Monticello to project The Trunk: http://source.squeak.org/trunk/Monticello-bf.604.mcz ==================== Summary ==================== Name: Monticello-bf.604 Author: bf Time: 24 November 2014, 6:30:56.448 pm UUID: b1f76c5b-7165-4dcb-bc69-07a1c7978fd7 Ancestors: Monticello-mtf.603 Deal with non-class globals. =============== Diff against Monticello-mtf.603 =============== Item was changed: ----- Method: MCMethodDefinition>>actualClass (in category 'accessing') ----- actualClass + ^Smalltalk at: className ifPresent: [:class | + class isBehavior ifTrue: [classIsMeta ifTrue: [class classSide] ifFalse: [class]]]! - ^Smalltalk at: className - ifPresent: [:class | classIsMeta ifTrue: [class classSide] ifFalse: [class]]!

1 0

New Cog VMs available
by Eliot Miranda 24 Nov '14

24 Nov '14

... at http://www.mirandabanda.org/files/Cog/VM/VM.r3144. Apologies; the 3143 VMs were broken. CogVM binaries as per VMMaker.oscog-eem.950/r3144 Fix regression in new primitiveBitShift. Introduce >>> as an explicitly signed shift. 64-bit ize magnitude64BitIntegerFor:neg:, positive64BitIntegerFor: & signed64BitIntegerFor: -- best, Eliot

2 2

Re: SqueakJS
by Bert Freudenberg 23 Nov '14

23 Nov '14

> On 23.11.2014, at 05:51, Jerry Bell <jdbellomm(a)gmail.com> wrote: > > Hi Bert, > > That shouldn’t have been necessary … I specifically wrote it using the original syntax so it would work even in old images where the Compiler does not know about named primitives yet. What exactly didn’t work? > > If I use the original syntax, I just get an "Error: a primitive has failed" error. I have no idea what might cause it. Ah, the compiler now optimizes away the unused literal array: (JS compiledMethodAt: #doesNotUnderstand:) literals ==> #(#primitiveFailed) whereas the old Squeak compiler leaves it in: (JS compiledMethodAt: #doesNotUnderstand:) literals ==> #((JavaScriptPlugin primitiveDoUnderstand 0 0) primitiveFailed) The VM looks at the first literal to find the primitive. But I found a way to fool the compiler, which also makes the JS bridge work in current Squeak. > MVC strongly assumes that it has access to the pixels on the display. For example, if you open a menu, it renders the outline and text items on the display. Then, to highlight an item, it does not re-render it, but draws a rectangle in xor-mode, reversing all pixels. This is very efficient … provided you have direct access to pixels. Which you don’t have on a canvas - reading back pixels is slow. Canvas wants to be treated as a write-only surface. > > > OK, that is good to know. Does Morphic have the same assumptions? Conceptually it doesn’t, all drawing operates on a Canvas (an abstract class). However, the concrete class has always been FormCanvas, which does expose the underlying bitmap. There is code that makes use of that. But it shouldn’t be too hard to clean that up. E.g., there is also a PostScriptCanvas that can be used to generate a vector rendering. But it never has been used interactively, it’s just for printing. But making a JSCanvas subclass would be a reasonable course of action. Older versions of Morphic are quite zippy compared to more recent ones (e.g. the Scratch image uses an old Morphic). > You go and build the ultimate run-everywhere object engine, and then you want to go and clutter it up with HTML? :) As I said, it really depends on your goals :) My initial goal for SqueakJS was just to explore how far we can get in making existing Squeak images work. Which is why I only now added the JS bridge, because if you use it, that code won’t run anywhere else (not until we implement the JavaScriptPlugin for other VMs anyway, which is certainly possible by embedding v8/node, but not trivial, and then your code might still assume being run in a browser). > My questions on Canvas accelerated Squeak graphics are just random thoughts right now, before I go and start real work on my first real Squeak MVC web app. I am by no means a web expert, but whenever I am trying to build web applications it feels like I'm dealing with a tottering tower of kludges. I think HTML may be fine for building a web page, but Canvas is the future for interactive web applications. I don't want to spend my time trying to trick CSS into doing what I want, or trying to compensate for various subtle differences between browsers. I just want to have the same complete control over what my web application does as I would if I were building a desktop application. I think Canvas has the potential to provide that. Sounds cool! > But, Canvas is pretty low level, you can draw pretty text but text layout is up to the programmer - and there aren't many good Canvas frameworks out there yet. Most assume you want to build a game. There are a few exceptions, for example ZebKit looks pretty nice: http://www.zebkit.com <http://www.zebkit.com/>, and SpreadJS is a nice fast spreadsheet component implemented using Canvas: http://wijmo.com/products/spreadjs/ <http://wijmo.com/products/spreadjs/> - I've used SpreadJS in a UI with a few hundred thousand cells and it feels like a desktop app. > > I considered building a version of Morphic in Amber, but I didn't get too far- it was going to be a lot of work. I had pretty much decided to build my next app in pure JavaScript using ZebKit. And then I saw SqueakJS... Squeak in Canvas, easy to take advantage of existing JavaScript libraries, and fast enough to use for a real-world web app on an iPad (at least in MVC). Good enough for me for now! > > A couple more observations: one of the 'killer features' of Dart when running in a Dart VM is object memory snapshots, to speed up application loading - you already have this in SqueakJS using regular old JavaScript. Well, on startup SqueakJS still has to create a JavaScript object for each Smalltalk object, which takes more than a second for large images). I wonder … we could actually store the ready-made objects in IndexedDB … which would give us extremely fast image loading and saving. Now there’s an idea! > Also, SpreadJS is nice, but un-minified it weighs in at 6.39MB, and minified it is 1.83MB. The entire minimal-MVC image is 1.61MB. I don't think download sizes will be too much of an issue for SqueakJS web apps! Not sure if you have seen it, but the Makefile in the run/ directory combines VM and generated plugins into a single file (squeak-all.js, 1.1 MB). It minifies to 0.4MB and gzipping that results in just 0.1 MB :) > In summary, I don't really need any of the fancy Canvas accelerated graphics right now. But, I think SqueakJS has a lot of potential, far beyond being just a way to run Squeak images in a web browser. It seems to me that something pretty close to SqueakJS could be a lot like the Web should have been from the beginning! Indeed! This is pretty much why SqueakJS exists at all - as an experiment to build something using the Web as a universal runtime environment. If you take that experiment and make it do something useful, awesome! Btw, for discussion not directly related to the VM, but how to build a system on top of it, we might rather want to move off vm-dev and continue on squeak-dev. CC’ed. - Bert -

1 0

Daily Commit Log
by commits＠source.squeak.org 23 Nov '14

23 Nov '14

Changes to Trunk (http://source.squeak.org/trunk.html) in the last 24 hours: http://lists.squeakfoundation.org/pipermail/packages/2014-November/007380.h… Name: CollectionsTests-eem.224 Ancestors: CollectionsTests-nice.223 Use isFloat instead of class == Float ============================================= http://lists.squeakfoundation.org/pipermail/packages/2014-November/007381.h… Name: KernelTests-eem.282 Ancestors: KernelTests-dtl.281 Use isFloat instead of class == Float =============================================

1 0

Separator character in class names
by Trygve Reenskaug 22 Nov '14

22 Nov '14

Hi, I'm relying on decomposing composite class names. An example class name is /BB5aBank/ I decompose it into /#(BB5a Bank)/ but am using ugly and error-prone hacks to achieve it. It would be very helpful if I could use some kind of separator in the class name, e.g., /BB5a:Bank/ I've tried brute force by renaming class /BB5aBank /to /BB5a:Bank/ This is permitted and Smalltalk points to the renamed class. Compiling a method that is using the new class name fails: /bank := /Expression expected ->/BB5a:Bank new/ I've looked into the Parser code, but got lost. Anybody have any ideas? --Trygve

2 3

The Trunk: KernelTests-eem.282.mcz
by commits＠source.squeak.org 22 Nov '14

22 Nov '14

Eliot Miranda uploaded a new version of KernelTests to project The Trunk: http://source.squeak.org/trunk/KernelTests-eem.282.mcz ==================== Summary ==================== Name: KernelTests-eem.282 Author: eem Time: 22 November 2014, 7:59:02.387 am UUID: ba4a5d95-b7b2-48b8-a482-bc6d3262016b Ancestors: KernelTests-dtl.281 Use isFloat instead of class == Float =============== Diff against KernelTests-dtl.281 =============== Item was changed: ----- Method: IntegerTest>>testBigReceiverInexactNthRoot (in category 'tests - mathematical functions') ----- testBigReceiverInexactNthRoot " IntegerTest new testBigReceiverInexactNthRoot " "Inexact 3rd root (not a whole cube number), so a Float must be answered. However, receiver is too big for Float arithmethic." | bigNum result | bigNum := (100 factorial raisedTo: 3) + 1. "Add 1 so it is not a whole cube" self assert: bigNum asFloat isInfinite. "Otherwise, we chose a bad sample" result := bigNum nthRoot: 3. + self assert: result isFloat. - self assert: result class == Float. self deny: result isInfinite. self assert: result = 100 factorial asFloat. "No other float is closer. See following line" self assert: 100 factorial asFloat = (100 factorial+1) asFloat! Item was changed: ----- Method: IntegerTest>>testBigReceiverInexactSqrt (in category 'tests - mathematical functions') ----- testBigReceiverInexactSqrt " IntegerTest new testBigReceiverInexactSqrt " "Inexact 3rd root (not a whole cube number), so a Float must be answered. However, receiver is too big for Float arithmethic." | bigNum result | bigNum := 100 factorial squared + 1. "Add 1 so it is not a whole square" self assert: bigNum asFloat isInfinite. "Otherwise, we chose a bad sample" result := bigNum sqrt. + self assert: result isFloat. - self assert: result class == Float. self deny: result isInfinite. self assert: result = 100 factorial asFloat. "No other float is closer. See following lines" self assert: (result successor asFraction squared - bigNum) abs >= (result asFraction squared - bigNum) abs. + self assert: (result predecessor asFraction squared - bigNum) abs >= (result asFraction squared - bigNum) abs! - self assert: (result predecessor asFraction squared - bigNum) abs >= (result asFraction squared - bigNum) abs.! Item was changed: ----- Method: ScaledDecimalTest>>testInexactNthRoot (in category 'tests') ----- testInexactNthRoot | tenth cubicRoot3 fifthRootTenth three | three := 3.0s1. cubicRoot3 := three raisedTo: 1/3. + self assert: cubicRoot3 isFloat. - self assert: cubicRoot3 class = Float. self deny: cubicRoot3 squared = 3. tenth := 0.10s2. fifthRootTenth := tenth raisedTo: 1/5. + self assert: fifthRootTenth isFloat. + self deny: fifthRootTenth squared = tenth! - self assert: fifthRootTenth class = Float. - self deny: fifthRootTenth squared = tenth.! Item was changed: ----- Method: ScaledDecimalTest>>testInexactSqrt (in category 'tests') ----- testInexactSqrt | tenth sqrt3 sqrtTenth three | three := 3.0s1. sqrt3 := three sqrt. + self assert: sqrt3 isFloat. - self assert: sqrt3 class = Float. self deny: sqrt3 squared = 3. tenth := 0.10s2. sqrtTenth := tenth sqrt. + self assert: sqrtTenth isFloat. + self deny: sqrtTenth squared = tenth! - self assert: sqrtTenth class = Float. - self deny: sqrtTenth squared = tenth.!

1 0

The Trunk: CollectionsTests-eem.224.mcz
by commits＠source.squeak.org 22 Nov '14

22 Nov '14

Eliot Miranda uploaded a new version of CollectionsTests to project The Trunk: http://source.squeak.org/trunk/CollectionsTests-eem.224.mcz ==================== Summary ==================== Name: CollectionsTests-eem.224 Author: eem Time: 22 November 2014, 7:57:15.254 am UUID: 31b12501-61bb-42ae-a8c5-3e42fd0ff717 Ancestors: CollectionsTests-nice.223 Use isFloat instead of class == Float =============== Diff against CollectionsTests-nice.223 =============== Item was changed: ----- Method: SortedCollectionTest>>testReversed (in category 'basic') ----- testReversed | sc1 sc2 sc3 | sc1 := #(1 2 3 4) asSortedCollection. self assert: sc1 reversed asArray = sc1 asArray reversed. self assert: sc1 reversed class = SortedCollection description: 'reversing a SortedCollection should answer a SortedCollection'. sc1 removeFirst; removeLast. sc2 := sc1 reversed. self assert: sc2 reversed asArray = sc1 asArray. sc2 add: 3/2; add: 1/2; add: 7/2. self assert: sc2 asArray = {7/2. 3. 2. 3/2. 1/2}. sc3 := #(1 2 3 3.0 4) asSortedCollection. self assert: sc3 reversed asArray = #(4 3.0 3 2 1). + self assert: (sc3 reversed at: 2) isFloat! - self assert: (sc3 reversed at: 2) class = Float. - !

1 0

New Cog VMs available
by Eliot Miranda 21 Nov '14

21 Nov '14

... at http://www.mirandabanda.org/files/Cog/VM/VM.r3143. CogVM binaries as per VMMaker.oscog-tpr.949/r3143 Spur: Fix regression in primitiveNewWithArg from VMMaker.oscog-eem.859 which caused basicNew: on fixed classes to raise OutOfMemory instead of the relevant error. Have checkForAndFollowForwardedPrimitiveState follow other references in the frame if a forwarder is found on the stack. All: Add a first cut at the primitiveSlotAt[Put] primitives. Reimplement primitiveBitShift for 64-bits. Define areIntegers:and: for 64-bit Spur, and move it into the ObjectMemory hierarchies. Add maxSmallInteger and minSmallInteger and matching manifest constants. Use these in LargeIntegersPlugin and Matrix2x3Plugin. Use numBytesOf: instead of lengthOf: in the 64-bit integer conversion routines. Reorder cases in Spur's numBytesOf: to put common case first. -- best, Eliot

3 5

[Fwd: Re: [Pharo-dev] Float hierarchy for 64-bit Spur]
by Ben Coman 21 Nov '14

21 Nov '14

(sorry, I didn't notice the cross-post) Eliot Miranda wrote: > Hi All, > > 64-bit Spur can usefully provide an immediate float, a 61-bit subset > of the ieee double precision float. I wonder if class SmallDouble would be more intention revealing? In practice 61 bits will be "more than enough"(tm) for anyone. But I can envisage in a business environment environment software needing to comply with (sometimes irrelevant) feature checklists, with one of those likely being full 64 bit compliant IEEE Doubles. Can we have such a class, to which 61 bit floats are auto-promoted as required? > The scheme steals bits from the > mantissa to use for the immediate's 3-bit tag pattern. So values have > the same precision as ieee doubles, but can only represent the subset > with exponents between 10^-38 and 10^38, the single-precision range. > The issue here is how to organize the class hierarchy. > > The approach that looks best to me is to modify class Float to be an > abstract class, and add two subclasses, BoxedFloat and SmallFloat, such > that existing boxed instances of Float outside the SmallFloat range will > become instances of BoxedFloat and instances within that range will be > replaced by references to the relevant SmallFloat. > My first few pages of search results lead to a few references in conversation, but nothing that described what a boxed float is. Can someone explain? btw, http://www.ctan.org/pkg/float also mentioned boxed float, ruled float and plaintop float Anyone familiar with those? > With this approach ... > > - Float pi etc can still be used, even though they will answer instances > of SmallFloat. But tests such as "self assert: result class == Float." > will need to be rewritten to e.g. "self assert: result isFloat". This is probably a good change anyway. > > - BoxedFloat and SmallFloat will not be mentioned much at all since > floats print themselves literally, and so the fact that the classes have > changed won't be obvious. > > - the boxed Float primitives (receiver is a boxed float) live in > BoxedFloat and the immediate ones live in SmallFloat. Making SmallFloat > a subclass of Float** poses problems for all the primitives that do a > super send to retry, since the boxed Float prims will be above the > unboxed ones and so the boxed ones would have to test for an immediate > receiver. ** do you mean the current Float, or after Float become abstract? > > > An alternative, that VW took (because it has both Float and Double) is > to add a superclass, e.g. LimitedPrecisionReal, move most of the methods > into it, and keep Float as Float, and add SmallFloat as a subclass of > LimitedPrecisionReal. Then while class-side methods such as pi would > likely be implemented in LimitedPrecisionReal class, sends to Float to > access them find them via inheritance. An automatic reorganization > which moves only primitives out of LimitedPrecisionReal is easy to write. > A Float is defined as a limited precision real [1] having several types of precision, so I like the first option. [1] http://en.wikipedia.org/wiki/Floating_point cheers -ben

3 6

← Newer
1
2
3
4
5
6
7
8
...
13
Older →