- Pharo-users - lists.pharo.org

FFI and when to convert the stack pointer (an Alien) into an ExternalAddress
by Holger Freyther April 4, 2019

April 4, 2019

I am toying with Ben's bindings to clang-c (C interface to the Clang C-indexer API). One of the main features of the C API is a way to visit the translation unit. This is implemented as a callback into Smalltalk with with "cursors" providing indirect access and simplified access to the node of the AST. The return result indicates how to continue the traversal (continue, stop, recurse). FFICallback signature: #( CXChildVisitResult ( CXCursor cursor, CXCursor parent, CXString client_data)) block: [ :cursor :parent :clientData | cursor spelling. <- booom but no crash aCXChildVisitResult value "@1" ]. "cursor" will be an Instance of CXCursor and is passed by value and the C struct size is >16 bytes (on Unix this is passed on the stack). The resulting CXCursor has a handle that is invalid to perform any call-out with. CXCursor getHandle FFIExternalStructureReferenceHandle handle: an Alien (isPointer = true, the stack ptr offset: 0 ExternalStructure defines valid types as ByteArray (the data) or ExternalAddress (the pointer). FFIExternalStructureReferenceHandle is missing there and I will create a separate thread/ticket about it. So somewhere from: FFICallback>>#valueWithContext:sp: ... FFICallbackArg>>#extractStructType: FFIExternalType>>#handle:at: FFIExternalStructureType>>#basicHandle:at: ee will need to convert the handle + offset (one based) to an ExternalAddress. Now the protocol of >>#handle:at: and >>#integerAt:, >>#pointerAt: indicate that we want to carry address+offset into late of the conversion. And I can't point my finger into where it should be done. I have created: Alien>>#referenceStructAt: byteOffset length: length ^self isPointer ifFalse: [super referenceStructAt: byteOffset length: length ] ifTrue: [ (ExternalAddress fromAddress: self addressField) + (byteOffset - 1)] This "fixes" it. But that is at the last point in the conversion. It shadows another problem as well. So in fact maybe changing: FFICallbackArgumentReader>>#extractPointerOfType: and >>#extractStructType: to read: ... pair := self nextBaseAddressForStructure: type. baseAddressToRead := ExternalAddress fromAddress: pair first address. <- modified offsetOfBaseAddress := pair second. ... is this the right way forward? We convert the stack to an ExternalAddress early on and use it. With the above change to FFICallbackArgumentReader the CXCursor from the stack will read as: CXCursor getHandle FFIExternalStructureReferenceHandle handle: ExternalAddress offset: 0 I am still not able to push this cursor to the stack. But that is for another mail. WDUT? holger

1 0

Re: [Pharo-users] Comment on: Pharo performance issues
by Marcus Denker April 3, 2019

April 3, 2019

> > I spent quite some time with Alexandre Bergel's NeuralNetwork implementation and stumbled across many implementations that cause general performance issues. > > While, I "fixed/altered" only those parts affecting my code, I would offer to search trough the image an alter the parts similar to those already altered. > > In such case that somebody would like to move a greater amount of code towards review and integration into current development how would the process of such contribution look like? > > The regular path of reporting bugs, or just forking on Github and pull request? The best is to first report an issue, then do a pull request. There is a pointer to a tutorial here: https://github.com/pharo-project/pharo/blob/Pharo8.0/CONTRIBUTING.md <https://github.com/pharo-project/pharo/blob/Pharo8.0/CONTRIBUTING.md> (If you get stuck somewhere, please tellâ¦ I think it might needs some update for Pharo8, it is on my TODO to check it carefully) > > I am extremely busy right now, but during the nights I could slowly attempt to work some performance issues out. > That would be great! Marcus

1 0

Comment on: Pharo performance issues
by Sebastian Heidbrink April 2, 2019

April 2, 2019

Hi! A friend of mine just just pointed this entry out to me https://pharoweekly.wordpress.com/2019/04/02/about-performance-issues/ I spent quite some time with Alexandre Bergel's NeuralNetwork implementation and stumbled across many implementations that cause general performance issues. While, I "fixed/altered" only those parts affecting my code, I would offer to search trough the image an alter the parts similar to those already altered. In such case that somebody would like to move a greater amount of code towards review and integration into current development how would the process of such contribution look like? The regular path of reporting bugs, or just forking on Github and pull request? I am extremely busy right now, but during the nights I could slowly attempt to work some performance issues out. Kind wishes, Sebastian

1 0

[ANN] Pharo Workshops in Switzerland: April 30 to May 3rd
by Marcus Denker April 2, 2019

April 2, 2019

Hi, Kertyx is organising a Workshop Series, the Digital Healthcare Innovation Workshops Part of this series will be a workshop around Pharo: Interactive and efficient ways to build technology stacks for experienced developers. - Where? Lavigny/Switzerland - When? April 30 to May 3 - Event web site: http://kertyx.events â¢ Workshop 1 â Introduction to Pharo (Marcus Denker) â¢ Workshop 2 â Web development using Seaside (Cyril Ferlicot) â¢ Workshop 3 â New Integrated Development Environment (IDE) â Glamorous Toolkit (Tudor Girba) Registration via eventbrite is available for all 3 workshops independently at http://kertyx.events Marcus

1 0

Google Protobuf - Small update and parsing a tensorflow GraphDef
by Holger Freyther April 2, 2019

April 2, 2019

A short update on the current state of my implementation. I heard there is interest to use it to parse tensor flow models. 0.) Load the BaselineOfProtobuf from https://github.com/zecke/pharo-protobuf 1.) Generate code Use the Google protoc to generate a descriptor set: $ protoc -o tf.pb --include_imports tensorflow/core/framework/graph.proto And then use this descriptor to generate code: | descriptor nameTable generator | descriptor := GPBFileDescriptorSet materializeFrom: 'tf.pb' asFileReference binaryReadStream. nameTable := GPBTypeNamesVisitor new. nameTable customPrefix: 'TF_'. generator := GPBGeneratingVisitor new typeNames: nameTable; targetPackage: 'Tensorflow-Definitions'. descriptor visit: nameTable. generator visit: descriptor. 2.) Parse a model (e.g. the inception v3 model) TF_GraphDef materializeFrom: 'inception_v3_2016_08_28_frozen.pb' asFileReference binaryReadStream 3.) ??? I guess load it into tensorflow? I am not sure if the endianness for the Float is correct (if not the weights are wrong so please be careful and have a look). There are still plenty of TODOs left. JSON and TextProto parsing needs to be implemented. Working on the official regression suite is needed as well. Strict/Non-strict modes for parsing are needed as well.

1 0

How are you supposed to recover code in Pharo - epicea seems half finished?
by Tim Mackinnon April 1, 2019

April 1, 2019

Hi - Iâve been hit by this a few times now, and Iâm scratching my head how its supposed to work. I moved some classes to another package and tag (but sometimes Pharo does this badly and rather than a sub-tag it makes a new top level package with a hyphen in it). So I then committed my changes to git, and thought nothing of it - but the flaw here is that with a new top package my classes didnât get commited!!! I then changed branch and started something else, but later saw this top level package and thought it was an Iceberg blip (it sometimes doesnât reload packages properly when changing branches) - and so I deleted those classes. A few minutes later it suddenly - it dawned on me Oh noâ¦.. Ok fine - we have changes tech right? But with no .changes file, we use epicea ????? But hereâs the frustration - I know the classes, but they are thousands of lines down the changes list now (Iâve changed branch and loaded new code). Where is the filter by specific class? And gosh its darn slowâ¦. Iâm missing the text editable .changes file right now as this doesnât work so well to get to a specific place to recover methods, comments etc. Am I doing something wrong? Missing something obvious? Tim

1 0

why is asDictonary a class method
by Roelof Wobben March 31, 2019

March 31, 2019

Hello, Im busy with a new challenge from exercism. Where I have to keep track of a robot , where it facing and on that coordinate the robot is. so I made this function what the test wanted createDirection: aString position: aCollection Â Â Â self robot: (Robot directionLooking: aString) yourself. Â Â Â ^ self robot asDictonary I can see that on the first part a new robot is made with the right data. but the test wants the data back as a Dictonary that is why I made the self robot asDictonary line but to my suprise the compiler wants it be a class method where I expect it to be a instance method. Can someone explain to my why this is ? Roelof

4 6

How to generically find methods in class hierarchies? (Aka narrow spotter?)
by Tim Mackinnon March 31, 2019

March 31, 2019

Iâm wondering if there are better ways of finding things that I am missing. In another thread I was asking about how to wrap the lookup of an item at an index in a collection. I had browsed the collection hierarchy and not seen anything (I kept coming back to the fact that I dont find the protocol categories very helpful, particularly in the collection hierarchy). I had also used spotter - and to explain - if you type âwrapâ into spotter there are 400 methods that come up, and when you scroll down them, after about 100 the list starts to go blank (bug?). Spotter seems like the key however, but Iâm wondering if there is a way to search for âwrapâ methods in a hierarchy (limit it, and not search the full image)? I tried typing Collection, diving in and typing wrap - but that doesnât work. However I thought I had read that there was a way to search in a hierarchy - but I can seem to work it out? (Also - while the show inherited methods in Calypso is useful - I find itâs not great when you have half a concept in mind as depending on prefixes and some of the less common naming of things, it can be hard to find stuff when itâs alphabetical and not filtered. So is there a trick Iâm missing? Tim Ps - Peter spotter the method I was after was called atWrap:, so it doesnât begging with w, and hence I missed it. Sent from my iPhone

3 5

Easiest way to wrap around the indexed value in a list (mod doesn't quite work)
by Tim Mackinnon March 30, 2019

March 30, 2019

Hey guys - Iâm wondering if someone might have a nice way of selecting the previous value in a list - but wrapping around to the last value if you are at the front. Essentially I have an exercise with some values in a collection like: list := #($a $e $o $u). And I have an index into the list, and want to loop around from index 1, back to 4 when I hit the beginning of the list. While moving forwards is pretty easy with mod - e.g. ^list at: (index \\ list size + 1) Iâm struggling with the cleanest way to do this backwards - as the following is unreadable to me (assuming index = 1) ^list at: (index - 2 \\ list size + 1) And then I tried (which isn't bad - but still not quite right). ^list before: (list at: index) ifAbsent: [ids last]. Iâm sure there is a better way to do this - but itâs alluding me? Tim

8 22

Resolving DNS in-image
by Holger Freyther March 29, 2019

March 29, 2019

Norbert and me looked at using DNS for service discovery and ran into some of the limitations of the NetNameResolver[1]. In the end I created an initial DNS implementation in Pharo called Paleo-DNS[2] to overcome these. DNS is a protocol we use every day but rarely think of. There is an active IETF community that is evolving the protocol and finding new usages (service discovery is one of them). In DNS there are different types of resource records (RR). The most commonly used ones in a client ("stub") are "A" for IPv4 addresses, "AAAA" for IPv6 addresses, "CNAME" for aliases, "SRV" records. So far only support for "A" records was implemented. So if you are curious about DNS then this is a great opportunity to add your favorite RR implementation to it and send a PR. There are probably 20+ of them to go. ;) Query example using DNS-over-TLS (DoT) to Google Public DNS PaleoDNSTLSTransport new destAddress: #[8 8 4 4] port: 853; timeout: 2 seconds; query: (PaleoDNSQuery new transactionId: (SharedRandom globalGenerator nextInt: 65535); addQuestion: (PaleoRRA new rr_name: 'pharo.org.'); addAdditional: (PaleoRROpt new udpPayloadSize: 4096)) [1] It's blocking on Unix, on Mac only one look-up may occur at a time and it returns exactly one address. There is also no IPv6 support. [2] https://github.com/zecke/paleo-dns

4 6