Search
Talks
Message in a Bottle
Edit
#ruby-implementations
#method-dispatch
#jruby
#just-in-time-jit
#ruby-internals
#performance-optimization
Summarized using AI

Message in a Bottle

Konstantin Haase • September 29, 2011 • New Orleans, Louisiana • Talk

Summary of "Message in a Bottle" by Konstantin Haase

In this talk delivered at RubyConf 2011, Konstantin Haase explores the intricate workings of method dispatch in Ruby implementations, namely MRI (Matz's Ruby Interpreter), JRuby, and Rubinius. The metaphor of a "message in a bottle" symbolizes method calls to objects, encapsulating the central theme of method dispatch and its internal mechanisms across different Ruby environments.

Haase begins by acknowledging his sponsors and providing a brief background on his experience, including his involvement with Sinatra and Ruby core development. He points out the existence of numerous Ruby implementations and narrows the focus to the three major ones.

Key points covered in the presentation include:
- Method Dispatch Overview: Method dispatch involves searching for and executing methods efficiently. Haase explains that each implementation uses different techniques to enhance speed.
- Inline Caches and Lookup Tables: These mechanisms are essential for speeding up method searches. Inline caching allows the runtime to store the location of the most recently called method, minimizing lookup time during subsequent calls.
- Execution and Bytecode: The talk highlights that all three implementations compile Ruby code to bytecode, which is central to method execution. Haase delves into how different implementations process this bytecode, emphasizing inline caching strategies used by Rubinius and JIT (Just-In-Time) compilation.
- InvokeDynamic in JRuby: A major feature of JRuby is its use of invokeDynamic, allowing dynamic method dispatch. This allows improved performance by creating flexible method call sites that are optimized at runtime.
- Coding Examples: Throughout the talk, Haase provides several coding examples to illustrate the dispatch mechanisms and the differences between the implementations, emphasizing how performance can be improved with various optimization techniques.
- Challenges with MRI: He notes the general slowness of MRI due to its interpreted nature and discusses potential avenues for improvement, such as super instructions and better JIT compilation strategies.
- Discussion and Q&A: Towards the end, the audience engages with Haase, asking about various details of Ruby's method dispatch, inline caches, and implementation specifics, reflecting the complex nature of Ruby's execution model.

Conclusion

Haas concludes that understanding the internals of these Ruby implementations is crucial for developers looking to optimize their code. He encourages the community to explore these concepts and contribute to improving Ruby's performance across various platforms.

The talk concludes with an acknowledgment of the collaborative nature of Ruby development and the importance of community contributions to enhance the language’s capabilities.

Message in a Bottle
Konstantin Haase • New Orleans, Louisiana • Talk

Date: September 29, 2011
Published: December 12, 2011
Announced: unknown

What does really happen when we call a method? How do the different Ruby implementations actually figure out what code to execute? What plumbing is going on under the hood to get a speedy dispatch? In this talk we will have a look at the internals of the four major Ruby implementations - 1.8, 1.9, JRuby and Rubinius, focusing on their dispatch. From look-up tables and call site caches, to inlining and what on earth is invokedynamic? Expect C, C++, Java, and, of course, Ruby code. But fear not, all will be explained!

RubyConf 2011

00:00:17.560 hi everybody um I'm Constantine I'm going to talk about message in a bottle but
00:00:24.720 before I start I wanted to thank engineer who are um
00:00:30.720 sponsoring um my talk here they're basically paying for a trip and toel so
00:00:36.079 without them I wouldn't be here right now so thanks engineered and yes as I said I'm
00:00:43.520 Constantine I'm on GitHub rch I have this extremely long Twitter
00:00:49.000 handle um uh it's probably easier to go to my
00:00:55.920 blog and click on Twitter than to type that in case you want to follow follow me or just check out my stream or
00:01:01.920 something um I'm currently a student I just graduated at that hustle pler
00:01:07.360 Institute down there that's in Germany near Berlin and I'm going back there to continue with my
00:01:13.880 master's program this year um I'm currently the maintainer of Sinatra
00:01:20.400 don't know if you guys have heard of it but it's quite handy and together with
00:01:26.159 um Ellen Harris I'm also working on a book about Sinatra which
00:01:31.759 is supposed to be released rather soon um I'm in the I'm a member of the r
00:01:40.240 core team and I'm currently working on robinus as in together with
00:01:49.240 Brian Ford from Portland and that's an internship sponsored Again by
00:01:56.799 engineer okay in this
00:02:02.079 talk I want to go into detail what's going on under the hood what makes Ruby
00:02:08.360 tick um how does how do those Ruby implementations actually run
00:02:16.280 Ruby um yes well here's here's the knowledge I'm
00:02:23.040 going to drop one8 is slow because it's interpreted and surprise
00:02:28.440 surprise uh so is one nine okay that was my talk
00:02:38.959 thanks I'm just kidding uh so Ruby
00:02:44.239 right um and message and Bottles um if you haven't figured it out
00:02:51.120 the message and the bottle is actually a metaphor for um sending a method call to
00:02:57.519 an object and that's what I'm going to talk about uh there will be a lot of code in this
00:03:03.560 presentation and I try to make sure that I don't lose anyone feel free to ask
00:03:08.799 questions right away rather than at the end
00:03:24.000 yes thanks um so for some reason there are a
00:03:29.879 lot of Ruby implementations out there I counted 58 or actually 56 because um or
00:03:36.560 maybe 55 depending on what you count as an
00:03:42.280 implementation um however I'm going to focus on the three large ones um J Ruby
00:03:49.720 MRI One n and rinus and here's a short overview I'll
00:03:56.760 talk about the method dis patch and execution were as while looking at at
00:04:02.400 MRI and I'm talking about inline crashes and jitting while talking about Rus and
00:04:08.760 I'll talk about invoke Dynamic while talking about J Ruby that doesn't mean that only those implementations do that
00:04:16.479 okay only J Ruby is doing invoke Dynamic but both rinas and J Ruby have a jit for
00:04:22.120 instance um I had some help for this talk from COI and a lot of help from
00:04:28.560 Brian and and also from Charles Nutter all those three guys are uh rather
00:04:35.240 active in the the specific Ruby implementations and all those three guys are here so buy them a beer or
00:04:42.720 something okay this is what I'm going to talk about
00:04:49.360 42 just for the example so um to make sure you all have an idea what I'm
00:04:55.039 talking about most of the examples will be talking about this code specifically
00:05:00.320 about what's happened at this point if we call this
00:05:06.080 method it's basically just uh um yeah checking if the value is 42 you can
00:05:14.400 pause in a string in the in the an integer well all the Ruby
00:05:19.720 implementations you're not supposed to understand that all that right now but um all the Ruby implementations compiled
00:05:26.360 to bite code and the basic idea of my talk is to to start from the bite code
00:05:31.560 and then look at what's going on there it's that's a bite code or somewhat the
00:05:38.360 bite code for um value TOS for MRI rubinus and J Ruby or J Ruby has you can
00:05:46.160 get different bite code but I come to that So the plan is if you want to
00:05:52.120 execute some Ruby code some method call you first have to search for the method
00:05:57.680 and then you have to execute the method but but you also want want to have that fast so how do we get that fast
00:06:04.639 well we search faster and we execute faster it's easy right okay we're done
00:06:11.240 um but how do we search faster um the first thing we can do is or one thing we
00:06:16.400 can do is inline caches I'll explain that which is basically in the code store where we're going um then look up
00:06:23.960 caches which is at the object store or at the class store where to find CL uh
00:06:30.120 methods that are inherited and we can do inlining to avoid a lookup or a method
00:06:35.759 called like at all and we can speed up the execution by
00:06:41.039 reducing the operations I think MRI is about to do that like instead of having one operation that does one thing have
00:06:49.039 one operation that does three things things so you don't have to have three op codes after each other you can do
00:06:55.720 just in time compilation to machine code um you can do inlining and well you can
00:07:02.479 speed up search cuz then your execution is faster because the main part of the execution is just looking for
00:07:09.440 methods all right MRI so if you call a method in MRI there
00:07:17.080 are different ways but basically almost always you end up in the C method called
00:07:24.080 rbor call zero uh that's the method and what is
00:07:32.120 important about this method is uh here's the search oh no I'm sorry
00:07:38.199 I'm on the wrong here's the search and here's the execute just like
00:07:43.400 I promised you uh okay now how does the search work
00:07:50.360 well uh did I say there's a lot of code in the slides it's basically the
00:07:57.479 um oh dang I'm I messed
00:08:06.280 up um it's basically the method dispatch we're all used to just start from the
00:08:12.199 class start from the class of the um of the object go through all the super
00:08:19.120 classes here and check if the method exists um well and super classes in that
00:08:27.159 case also includes included mod or I claes how they're called in MRI well and
00:08:34.039 that's a basic dispatch now that we have the method all we have to do is execute that
00:08:40.240 method um this is done by this method you don't have to get all that I just um
00:08:46.000 wanted to show you like I always put like comments where important stuff is
00:08:51.399 here it's actually going to execute or whatever the state of the of the current context is and what that does is an
00:09:00.160 extremely long method where there where I couldn't find important parts to tell you from but basically it goes to to uh
00:09:07.440 there's an instructions. def file where all the um
00:09:12.839 for every bite code there is a a chunk of code that will be executed when MRI
00:09:19.519 or Yar in that case is interpreting the B code and that looks like this oh I
00:09:26.000 always mess this up
00:09:32.320 I'm sorry I have the new show off with the present of VI and I'm not used to that um
00:09:39.959 so I would just wanted to show you the code that you get a feel of how the implementations look like this is
00:09:46.760 basically the code for the bite code for sending a method and and um it's taking
00:09:54.399 some um arguments from their s and um basically down here it's calling
00:10:01.640 the method which will again get you to RB to the method dispatch I just
00:10:09.240 show okay so there was a lot of code let's talk about
00:10:16.760 robinus well in robinus you have basically this
00:10:22.600 structure which is similar to what all the Ruby implementations have so every
00:10:27.959 module and the module is a super class of class and you know included module has a method table which is basically
00:10:33.959 like a hash um but more lowlevel construct you can easily easier access
00:10:39.800 from from C++ and um for every method there is a
00:10:45.480 bucket which is some metad data and it references to a compiled method which is
00:10:52.440 basically the B code representation of a method and the fun thing about robinus
00:10:59.000 is that you can actually do play with that from within Ruby and this is basically
00:11:07.000 the same lookup logic I showed you just implemented in in Ruby um note that we
00:11:13.519 have to use direct super class down there um to make sure we also
00:11:20.000 dispatching through all the included modules and if you get a hold of that
00:11:25.920 mod um method object that um compiled method object you can do some fun things
00:11:33.160 with that but first we have to get a hold of it with the lookup method and
00:11:38.720 then that will only give us the bucket so we have to ask for the compiled method and we'll well return a compiled
00:11:45.079 method object and we can decode that to take a look at the bite
00:11:50.560 code um we can also get the op codes which is more like the B the the decode
00:11:57.480 display is more like semantic names you can understand and this is
00:12:03.560 what is actually stored in memory or on file if you save the compiled output on
00:12:10.079 file and it's referencing literals in there I'll come to that on the next
00:12:15.760 slide that are stored in the literals Tuple Tuple is kind of like a fixed size
00:12:24.279 array um okay let's try to figure out what this does so in the literal Tuple
00:12:29.680 there's 2 s 42 and equal equal and we
00:12:35.000 also have a local that are used in there which is
00:12:40.240 value which in our case is 42 and then we have this Tuple of bytes that are the
00:12:47.639 method if you um if you take a close look at those we
00:12:52.800 start off with the B bite code for pushing a local onto the stack this back
00:12:58.639 here is the deck so we have 42 on the deck then this is the bite code for um
00:13:05.360 sending method which basically takes two arguments the first argument is the index of the method name in the literal
00:13:13.040 stuple up here to S has the index zero and the second one is the number of
00:13:18.680 arguments which is zero and then we have the string 42 on the stack after that we
00:13:25.680 push the literal with the index one onto the stack which is
00:13:32.800 42 um so we have 2 * 42 on the stack and then robinus has to duplicate that since
00:13:39.639 um if it doesn't and someone calls for instance G's a bang on that
00:13:45.320 string then um this will be modified if you
00:13:51.399 replace it for instance with 55 and next time you have to pass 55 to the method to get it to return true which actually
00:13:57.160 works if you go on GitHub on the repository I have a fun script there which modifies the
00:14:03.800 literals um and after that it's sending equal equal which has an a special bite code
00:14:11.199 so it can do some optimizations and we have to return since on bite Cod level there's no implicit
00:14:19.399 return okay but so we want to get inline caches into that so we don't have to
00:14:26.720 have a look up whenever it's doing the Cent de cuz you remember I told you I was talking about this 2 s
00:14:35.680 mainly so what rinus does it takes the bite code and replaces the 2s call with
00:14:42.240 an inline cache object um if you don't believe me here's
00:14:47.399 the code or at least the code which is using it as an inline cache object this is
00:14:54.320 basically again from the from the op codes defin from the instructions. D
00:15:00.399 from rinas and this is like I showed you before the send white code I mainly put
00:15:09.440 that in here because in my um Talk description I said I going to have C++ code and I made my slides and figured I
00:15:16.360 don't have C++ code so I figured let's put that slide in there C++ ladies and
00:15:25.279 gentlemen okay um uh the inline cash actually has a reference to something
00:15:32.920 called the VM method and the compiled method also has this reference so this
00:15:39.560 is the for instance the inline cache in this example um would be referencing
00:15:47.240 the VM method of 2s now the fun thing is um that the
00:15:53.519 inline Cache can actually during its lifetime change um
00:15:59.920 what method it is referring to or basically
00:16:05.480 anything oh yeah first of all the caching part is that that VM method is
00:16:10.639 already referencing the 2s code so it doesn't once the lookup is done and it's
00:16:16.360 in the inline cache it doesn't have to do the lookup again when visiting that
00:16:21.720 that point in the bite code unless um the type change so there's a
00:16:27.399 simple check where where it checks that nothing has changed if someone defined 2s then it will do the the lookup again
00:16:34.480 or if that's a different class it's that's coming in there it will do the look up again but if nothing changed
00:16:42.000 then it we use that method and yeah it can run that
00:16:49.199 method and as you can kind of see there you can specialize you can
00:16:56.279 change such a VM method um but why would you do that well
00:17:04.120 the standard version of such a method is just have the bite code stored in there and execute that bite code with the VM
00:17:11.760 basically interpret that bite code but that isn't fun yeah that isn't really fun except
00:17:18.799 for you don't have to do a lookup um you can actually have bite code with break
00:17:24.039 points in there which has the advantage of you can run a debugger and it will only slow down the method where you have
00:17:30.840 set a break point that's why rubinus can do debugging without slowing down the
00:17:37.200 the Ruby program which MRI can't um and you can specialize for
00:17:42.880 arguments like doing some optimization if that's a string you get in there
00:17:48.880 whatever that's currently under development I think and last but not
00:17:54.000 least you can jit that code that means you can actually instead of referencing bite code you will inter interpret you
00:17:59.720 can reference machine code that will well run directly on your
00:18:05.559 CPU and it works like this you take the uh robinus by code robinus translates
00:18:11.760 that into the immediate representation for llvm and then let's just llvm figure
00:18:18.280 out how to compile that to native code it is all the inlining optimization stuff that llvm is known for and that
00:18:25.600 way speeds up execution
00:18:30.919 all right any questions so far no
00:18:42.720 yes yes yes so the inline cache tracks how often it's called it tracks if
00:18:49.600 there's always the same coming in and um there's a heat and basically if
00:18:56.039 you call the method once then the heat values increase by one if the if the method does a loop then the uh heat
00:19:03.080 value is increased by one and I think it's something about if it hits 4,000 or
00:19:08.360 something uh then it will compile it to native code because basically if you
00:19:13.440 just execute the method once or twice it's way more expensive to compile it than to just interpret
00:19:20.559 it it's doing that at runtime just like Hotpot um okay and J Ruby is is
00:19:28.840 basically doing this something similar because that's um why it's using the jvm amongst
00:19:35.000 other things jvm is known for doing that well okay so let's talk about J Ruby and
00:19:42.360 lately if people talk about J Ruby and about performance and about method dispatch what they talk about is invoke
00:19:53.919 Dynamic okay but there's a lot of buzs about it and do you guys think you know
00:19:59.559 what it's doing no all right then let's try to figure
00:20:05.280 that out so there are two different kinds of B codes the K jvm trunk Master is skit
00:20:14.799 so it's Master uh the current Master can produce depending on what jvm version
00:20:19.840 you're on the one is the one you also get with the older J Ruby versions which is using invoke virtual and a custom
00:20:28.280 call object well in dyamic is using a custom call head object too
00:20:34.039 but um and the the new One is using um invoke Dynamic this is
00:20:43.240 again uh the extract from the method I showed you at the beginning that is calling
00:20:51.280 2s so let's so invoke virtual is the normal method bite code invocation for the JV M
00:20:59.120 and invoke Dynamic is the new hot thing they added um let's take a look at
00:21:05.840 those invoke virtual looks like this it's it uses the custom runtime call
00:21:11.919 site which actually encodes the dispatch Logic for uh J Ruby um and whenever you
00:21:19.840 come across that point will go to that call site and say Hey I want to call that J Ruby method and it
00:21:26.279 hence three arguments if you don't send any arguments the one is the thread
00:21:31.320 context that's running in the uh next is the caller and the third one is the
00:21:38.000 receiver and then it returns again a ruby object as you would expect
00:21:44.279 um the fun thing is about invoke Dynamic you could actually put any name you
00:21:51.080 want oh you don't see that at the where the call is and uh invoke Dynamic would
00:21:58.080 be able to um resolve that name or your logic would be able to do that but what
00:22:05.279 J Ruby is doing Instead at the moment I was told that that will change might change it's using
00:22:11.960 call just as an identifier whatever and
00:22:17.600 pushes the um the method name onto the stack so it
00:22:24.720 will actually be passed as argument to the method
00:22:30.520 and yeah again it Returns the a ruby object but with only that the jvm would
00:22:37.679 not know what to do so there's a signature down
00:22:42.919 here which looks like this fun okay so it's there is in J Ruby this
00:22:52.679 invocation Linker which is taking care of setting up invoke Dynamic properly
00:22:59.440 and the important thing about that Linker is it has um a bootstrap method
00:23:05.200 and a fallback method and it returns a call site and
00:23:10.919 that call site is like the inline cach in rinus it will be placed at that point
00:23:17.000 kind so it's kind of like that inline C and rinus it will be placed at that point in the bite code and how it's
00:23:24.840 supposed to be used is in in the in the bootstrap search for the method with
00:23:32.480 your method lookup logic and place it there and if if that fails or if a guard
00:23:39.640 fails you can place a guard in front of it then please call the fallback
00:23:44.960 method now the interesting thing about the bootstrap and J Ruby
00:23:50.960 is that it's not looking for the method like at all
00:23:56.000 because it doesn't know what argument will be passed as as the method name so
00:24:04.200 all it does is place a call site there that will on the first call call the
00:24:10.960 fallback so independent of whatever so there's no lookup logic in
00:24:18.360 here and then in the fallback there's actually the lookup logic
00:24:30.960 the lookup logic
00:24:36.880 is search with cache here and if it doesn't find it it has
00:24:43.399 actually it has to place a method missing in there and then it is actually updating the call site to point to that
00:24:52.159 method it has found and then calls the method so um
00:24:58.440 all the jvm optimizations can start from the second time you visit that point in
00:25:04.760 the bite code or then it probably happen whenever you place it there and then uh
00:25:10.760 the jvm can do all those inlining um hotspotting and so on way
00:25:17.120 easier because it's treated equally or somewhat equally to a
00:25:22.440 normal jvm method call oh
00:25:28.960 yes I kind of rushed through that did I
00:25:34.760 yes that was
00:25:40.799 it oh wow man I'm really sorry for rushing
00:25:47.080 through that cuz um I know I have this new presenter view from show off and it
00:25:54.600 indicates the color here and I was under the impression that I have time issues
00:26:00.240 okay should I explain something more in detail because I'm I was going through through a lot of code
00:26:07.240 in not that much time I think yes I have a specific question
00:26:13.360 about the G about um so the difference between
00:26:18.600 invoke and invoke Dynamic it sounds like what you said
00:26:23.720 invoke Dynamic does is the first time it tries to call a method it automatically nail and has to do a workout no that
00:26:31.120 yeah that's um so the first time it visits at IOD the bootstrap will be called to find um the method that should
00:26:39.880 should be placed there uh that it's going to the fall back is specific to the implementation J is using because
00:26:47.120 the issue is in the bootstrap you do not actually have aru the arguments you only
00:26:52.360 have the classes of the arguments past to it CU um the jvm as Java the lookup
00:26:59.279 and Method signatures that's all based on the number and the classes of the
00:27:04.480 arguments so that's what it uses as kind of key to fill up the caches and key for
00:27:10.760 doing inlining so within B Dynamic you basically have to map it onto those signatures for dis patches that do not
00:27:19.399 work in that way what what expires the caches are
00:27:24.440 there um so there are guards I think both robinus and jby are down to
00:27:30.559 actually one single check for normal method dispatch which is basically a
00:27:35.960 true false if it returns true then everything's still fine and if it returns false then we have to do it all
00:27:41.240 over again and it's basically like can number check something where if someone
00:27:47.840 defines a new method that that value changes and then other that value
00:27:53.240 changes for that class and then uh it will be invalidated
00:27:59.399 yes in your opinion what's the number one thing that MRI could add or
00:28:06.200 change um to do what become faster make
00:28:11.440 it fter yeah could add for instance so what MRI is doing is
00:28:19.880 trying to add um super instructions if you have like in the
00:28:25.679 white code
00:28:33.200 in the B code up here if you have um oh I'm always on one
00:28:39.679 slide if you have um get local and send then it has to go through the VM Loop
00:28:46.720 two times and what what they are trying to do is um for some cases for instance
00:28:52.480 in that case you could do a get local and send up code um you would only have
00:28:58.760 one up code and encode all the logic in there uh the issue with that is that um
00:29:04.799 you will not um you're not able to
00:29:10.000 change the class of complexity so if you if you
00:29:15.159 double the the SI the number of up codes the best thing you can do is cut the um
00:29:23.039 performance the the time it needs in half but that's not realistic so it will only scale the near it will not um
00:29:31.000 change the performance overall whereas um jitting and inlining can actually
00:29:36.240 change it over all because you enable a new much bigger world of optimizations
00:29:42.240 and even the CPU engineering people can help you run your stuff faster basically
00:29:48.279 so that would be one thing I think yes I saw on the rage slide that it was a
00:29:54.799 dment string yes that seems like potential to add a lot a lot of over do
00:30:01.159 you have any IDE okay so the thing is um
00:30:06.640 allation yes you you have to um Let me show that um because the string in the
00:30:14.159 in the literal to there actually one string and if you hand out the same string always and someone modifies it
00:30:20.559 then will be modified for all this it might be might be easier to understand if you take a look at
00:30:26.720 this um yes so I load the example code go to the
00:30:36.279 go to the literals tle of the method and place 55 in there and all of a sudden 55
00:30:42.200 is the ultimate answer and if you if you um Would Not Duplicate that string then
00:30:48.000 anyone could do uh take that string that it's handed to from the method and do um
00:30:55.240 replay pass 55 to it and then you could do really evil things by
00:31:13.000 accident so you don't have to do that with symbols cuz you can modify those
00:31:18.080 did that answer your question okay so try use symbols were possible uh yeah
00:31:24.519 but the um bad thing about symbols is that garbage C so if you use um for
00:31:31.519 instance symbols for user generated data or something then we'll just fill up
00:31:36.639 your memory and they'll never go
00:31:45.760 away feel free to ask
00:31:52.200 more okay you don't want to yes I think we have lot of time for
00:31:58.760 discussion why are symbols not garbage collected um cuz that's what ruby did
00:32:05.559 does um because symbols are supposed to be
00:32:10.760 um it's way cheaper if you and they're used for like the method handles for
00:32:15.840 interning and also um if they garbage collected then they would be created all
00:32:21.760 over um basically all the literals so if you if you define the method in rinus
00:32:29.159 for instance that string 42 is never garbage collected because it's referenced by that method so it will
00:32:34.880 only be garbage collected if the method is garbage collected and it would extremely complicate literals for
00:32:40.840 instance if you would garbage collect symbols but
00:32:46.200 um you might want to talk to don't oh yeah the big thing is the Syms the same
00:32:52.880 object everywhere so you have to have a reference to it somewhere you can't create a symbol Be
00:32:58.760 object yes some some languages some lists for example do have function
00:33:05.559 called uned so you can remove like user generated but it does not make sense to
00:33:11.639 do it
00:33:17.480 for yes there was another question back there I was I was nice meatball here just H all apart
00:33:24.880 but if somebody were interested in trying to understand and piece together
00:33:30.200 U or explore these virtual machines um like you've done in this study how would you how did you approach this
00:33:38.080 um I grabbed the hold of Brian Ford and told him explained that to me
00:33:44.679 fromus um I mean is there a particular say you know Class C file a starting
00:33:50.240 point um yes if you're in depends on
00:33:55.320 what you want to look at so um the both I found for myself that both
00:34:03.240 the rinus code and the J Ruby code is pretty easy to read but it's a ton of
00:34:09.399 code um if you if you go into the C Ruby code uh there are it depends on where
00:34:16.960 you go I would recommend I would recommend starting with the search method method because that's small code
00:34:23.839 it's pretty understandable all the related methods are right below it which is not always the case with the other
00:34:29.839 methods and then okay what I mean by how to understand this
00:34:39.040 um wow this is actually from the Wim
00:34:44.679 methods one of the so they all have pretty complex workflow and this doesn't
00:34:50.240 have the most complex workflow but still it has the pattern of uh having a
00:34:56.040 label in front of switch then that switch is nested there are more switches in that switch and from time to time it
00:35:02.800 says go to again and that is just might make sense
00:35:08.960 but it's really hard to understand if you just want to follow the the the
00:35:14.119 program flow and yeah I had the advantage of
00:35:20.320 just asking qu about that code something
00:35:28.520 yes and there's the book picture um any more questions yes over there again do you
00:35:35.240 think there's any chance or any possibility that the entire state of the execution stat could be serialized for
00:35:42.480 any of these I don't believe any implementation support um Meg does kind
00:35:49.280 of because um it's image based um apart from that no proper Ruby
00:35:57.000 implementation so supports that I'm not sure if that's hard or anything feel free to talk to all the people that are
00:36:03.079 here and make sure I'm with you cuz I I know that there have been discussion
00:36:10.119 about this from time to time but there's no implementation on any other apart from
00:36:18.000 me the Mega people are over there
00:36:25.520 okay um in that case thank you all for listening to my talk
Konstantin Haase
@rkh
Explore all talks recorded at RubyConf 2011
+55