Re: Dynamically binding to D code using extern(D)
On Thursday, 30 September 2021 at 22:30:30 UTC, jfondren wrote: 3. dynamic linking (option 2), performed arbitrarily at runtime, by your program. If linking fails, you can do whatever you want about that. That's actually "dynamic loading". https://en.wikipedia.org/wiki/Dynamic_loading
Re: Dynamically binding to D code using extern(D)
Okay, I do agree with you that I may have exaggerated with absolute intuitiveness, but I was talking about that intuitiveness for loading a symbol from a shared library. You're limited to using C's types - I think I don't understood what you meant with that, if the data type is known before head, it is possible to just declare it from the other side On Thursday, 30 September 2021 at 22:30:30 UTC, jfondren wrote: - you can't use overloading, lazy parameters, default values; you can't rely on scope parameters, etc., etc. - That seems to be pretty much more a problem for dynamically loading a function, although default values can be mirrored to in D API. - you can't casually hand over GC-allocated data and expect the other side to handle it right, or structs with lifetime functions that you expect to be called - That is another problem that doesn't seem related to the external linkage too, handling GC-allocated data with extern(D) doesn't stop it from it being garbage collected, I'm fixing that kind of error right now again. separate applications and some form of interprocess communication (pipes, unix sockets, TCP sockets) instead of function calls. - I'm pretty interested in how to make that thing work, but I think that would change a lot in how I'm designing my code, and with that way, it would probably become absolutely data oriented, right?
Re: Dynamically binding to D code using extern(D)
On Thursday, 30 September 2021 at 18:09:46 UTC, Hipreme wrote: I write this post as both a learning tool, a question and an inquiry. There are just a lot of drawbacks in trying to do function exporting while using D. The terms that people use are a bit sloppy. There are three kinds of 'linking' here: 1. static linking, performed during compilation, once. If linking fails, the compile files. 2. dynamic linking (option 1), performed when an executable starts up, before your program gains control, by the system linker. If linking fails, your program never gets control. 3. dynamic linking (option 2), performed arbitrarily at runtime, by your program. If linking fails, you can do whatever you want about that. All of the loadSymbol and 'userdata module' hassle that you're frustrated by is from option 2. Option 1 is really the normal way to link large shared libraries and there's nothing to it. What your code looks like that loads a shared library is just `import biglib;`, and the rest of the work is in dub, pkg-config, `LD_LIBRARY_PATH`, etc. Phobos is commonly linked in this way. Pretty much anything that isn't a plugin in a plugin directory can use option 1 instead of option 2. extern(C) advantages: - Code callable from any language as it is absolutely intuitive - Well documented You can call scalding water 'hot' even when you're fresh from observing a lava flow. People still find the C ABI frustrating in a lot of ways, and especially when they encounter it for the first time. But the C ABI rules the world right now, yes. The real advantages are - it 'never' changes - 'everyone' already makes it easy to use extern(C) disadvantages: - You will need to declare your function pointer as extern(C) or it will swap the arguments order. - you're limited to using C's types - you can't use overloading, lazy parameters, default values; you can't rely on scope parameters, etc., etc. - you can't casually hand over GC-allocated data and expect the other side to handle it right, or structs with lifetime functions that you expect to be called - very little of importance is statically checked: to use a C ABI right you need to very carefully read documentation that needs to exist to even know who is expected to clean up a pointer and how, how large buffers should be. (I wasn't feeling a lot of the C ABI's "absolute intuitiveness" when I was passing libpcre an ovector sized to the number of pairs I wanted back rather than the correct number of `pairs*3/2`) Option 2 dynamic linking of D libraries sounds pretty frustrating. Even with a plugin architecture, maybe I'd prefer just recompiling the application each time the plugins change to retain option 1 dynamic linking. Using a C ABI instead is a good idea if just to play nice with other languages. And if you were wanting something like untrusted plugins, a way to respond to a segfault in a plugin, like I think you mentioned in Discord, then I'd still suggest not linking at all but having separate applications and some form of interprocess communication (pipes, unix sockets, TCP sockets) instead of function calls. This is something that you could design, or with D's reflection, generate code for against the function calls you already have. But this is even more work that you'll have to do. If we add "a separate process telling you what to do with some kind of protocol" as a fourth kind of linking, then the respective effort is 1. free! it compiles, it's probably good! 2. free! if the program starts, it's probably good! 3. wow, why don't you just write your own loadSymbol DSL? 4. wow, why don't you just reimplement Erlang/OTP and call it std.distributed? maybe protobufs will be enough.
Dynamically binding to D code using extern(D)
I write this post as both a learning tool, a question and an inquiry. There are just a lot of drawbacks in trying to do function exporting while using D. That interface is absurdly confuse and that is probably why I've never seen a project here which made an use of extern(D) while using a DLL. While I'm making my DLL's generation, there are a lot of pitfalls that I can feel into. **Simple Function** ``` module something; extern(D) export int sum(int a, int b){return a + b;} ``` The correct way to bind to that function would be: ``` module app; import core.demangle int function(int a, int b) sum; void main() { sum = cast(typeof(sum))GetProcAddress(someDll, mangleFunc!(typeof(sum)("something.sum"); } ``` And that should be it for loading a simple function. Now, lets make our case a bit more complicated: **Overloaded function** ``` module something; extern(D) export int add(int a, int b) { return a + b; } extern(D) export float add(float a, float b) { return a+b; } ``` For loading those functions, the correct way would be ``` module app; import core.demangle; int function(int a, int b) sumInt; float function(float a, float b) sumFloat; int sum(int a, int b){return sumInt(a, b);} float sum(float a, float b){return sumFloat(a,b);} void main() { sumInt = cast(typeof(sumInt))GetProcAddress(dll, mangleFunc!(typeof(sumInt))("something.sum")); sumFloat = cast(typeof(sumFloat))GetProcAddress(dll, mangleFunc!(typeof(sumFloat))("something.sum")); } ``` Notice how much the overall complexity starts to increase as there seems to be no way to put get the overloads and there doesn't seem to be any advantage in using extern(D). **Static Methods** The only difference from the default functions is that we need to pass the class name as a module name. **Static Methods returning user data** That is mainly the reason I'm writing that post. It made me really wonder if I should really use extern(D). This section will use 3 files because after all, there is really a (consistency?) problem ``` module supertest; import ultratest; class SuperTest { extern(D) export static SuperTest getter(){return new SuperTest();} extern(D) export static UltraTest ultraGetter(){return new UltraTest();} import core.demangle; pragma(msg, mangleFunc!(typeof())("supertest.SuperTest.getter")); //Prints _D9supertest9SuperTest6getterFZCQBeQx pragma(msg, mangleFunc!(typeof())("supertest.SuperTest.ultraGetter")); //Prints _D9supertest9SuperTest11ultraGetterFZC9ultratest9UltraTest } ``` ``` module ultratest; class UltraTest{} ``` ``` module app; import core.demangle; void main() { //??? } ``` As you can see at module supertest, the pattern seems to break when returning user data for another module. From my knowledge, I don't know how could I get this function, specially because you will need to know: the module that you're importing the function + the module that where the userdata is defined for getting it. It seems pretty insane to work with that. extern(D) advantages: - extern(D) disadvantages: - Code only callable in D(probably no other language as a demangler) - I don't remember seeing any other code before in that post doing that, so, no documentation at all - You will need to call the demangler for binding to a symbol, which in my project, it could make each call to a unique type from the demangler costs 15KB - You will need to know the module which you imported your function - If your function returns userdata from another function, there doesn't seem to be any workaround - Doesn't provide any overloading binding support though the language has support to overloading extern(C) advantages: - Code callable from any language as it is absolutely intuitive - Well documented extern(C) disadvantages: - You will need to declare your function pointer as extern(C) or it will swap the arguments order. I have not even entered in the case where I tried overloading static methods, which I think it would need to declarate aliases to the static methods typings for actually generating a mangled name. I want to know if extern(D) is actually meant to not be touched. adr said that his use for that was actually when doing extern(C): //Funcs defined here extern(D): //Resets the linkage to the default one So, there are just too many disadvantages for doing extern(D) for binding it to any code, I would like to know where we can get more documentation than what I posted here right now (really, I've never saw any code binding to an extern(D) code). And I do believe that is the main reason why people usually don't use dynamic libs in D, it is just inviable as you would need to regenerate all the API yourself
Re: Rather Bizarre slow downs using Complex!float with avx (ldc).
On Thursday, 30 September 2021 at 16:40:03 UTC, james.p.leblanc wrote: D-Ers, I have been getting counterintuitive results on avx/no-avx timing experiments. This could be an template instantiation culling problem. If the compiler is able to determine that `Complex!float` is already instantiated (codegen) inside Phobos, then it may decide not to codegen it again when you are compiling your code with AVX+fastmath enabled. This could explain why you don't see improvement for `Complex!float`, but do see improvement with `Complex!double`. This does not explain the worse performance with AVX+fastmath vs without it. Generally, for performance issues like this you need to study assembly output (`--output-s`) or LLVM IR (`--output-ll`). First thing I would look out for is function inlining yes/no. cheers, Johan
Rather Bizarre slow downs using Complex!float with avx (ldc).
D-Ers, I have been getting counterintuitive results on avx/no-avx timing experiments. Storyline to date (notes at end): **Experiment #1)** Real float data type (i.e. non-complex numbers), speed comparison. a) moving from non-avx --> avx shows non-realistic speed up of 15-25 X. b) this is weird, but story continues ... **Experiment #2)** Real double data type (non-complex numbers), a) moving from non-avx --> avx again shows amazing gains, but the gains are about half of those seen in Experiment #1, so maybe this looks plausible? **Experiment #3)** Complex!float datatypes: a) now **going from non-avx to avx shows a serious performance LOSS** of 40% to breaking even at best. What is happening here? **Experiment #4)** Complex!double: a) non-avx --> avx shows performancegains again about 2X (so the gains appear to be reasonable). The main question I have is: **"What is going on with the Complex!float performance?"** One might expect floats to have a better perfomance than doubles as we saw with the real-value data (becuase of vector packaging, memory bandwidth, etc). But, **Complex!float shows MUCH WORSE avx performance than Complex!Double (by a factor of almost 4).** ```d //Table of Computation Times // // self math std math // explicit no-explicit explicit no-explicit // align alignalign align // 0.12 0.21 0.15 0.21 ; # Float with AVX // 3.23 3.24 3.30 3.22 ; # Float without AVX // 0.31 0.42 0.31 0.42 ; # Double with AVX // 3.25 3.24 3.24 3.27 ; # Double without AVX // 6.42 6.62 6.61 6.59 ; # Complex!float with AVX // 4.04 4.17 6.68 5.82 ; # Complex!float without AVX // 1.67 1.69 1.73 1.71 ; # Complex!double with AVX // 3.34 3.42 3.28 3.31# Complex!double without AVX ``` Notes: 1) Based on forum hints from ldc experts, I got good guidance on enabling avx ( i.e. compiling modules on command line, using --fast-math and -mcpu=haswell on command line). 2) From Mir-glas experts I received hints to try to implement own version of the complex math. (this is what the "self-math" column refers to). I understand that detail of the computations are not included here, (I can do that if there is interest, and if I figure out an effective way to present it in a forum.) But, I thought I might begin with a simple question, **"Is there some well-known issue that I am missing here". Have others been done this road as well?** Thanks for any and all input. Best Regards, James PS Sorry for the inelegant table ... I do not believe there is a way to include the beautiful bars charts on this forum. Please correct me if there is a way...)
Code coverage exit code 1 on failure?
What's the reasoning behind picking exit code 1 ? Makes it pretty much impossible to distinguish between a lack of coverage code 1 and a process code 1. Is there a handler where it can be overridden ?
Re: Why sometimes stacktraces are printed and sometimes not?
On Wednesday, 29 September 2021 at 12:15:30 UTC, Steven Schveighoffer wrote: On 9/29/21 6:57 AM, JN wrote: What makes the difference on whether a crash stacktrace gets printed or not? Sometimes I get a nice clean stacktrace with line numbers, sometimes all I get is "segmentation fault error -1265436346" (pseudo example) and I need to run under debugger to get the crash location. segmentation faults are memory access errors. It means you are accessing a memory address that is not valid for your application. If you are accessing the wrong memory, it means something is terribly wrong in your program. [...] So on Linux, I don't know the behavior on other OSs, the kernel sends SIGSEGV to your process which, if unhandled, simply terminates your program. It's an abnormal termination and thus the D runtime or whatever library that in a normal case takes care of printing the traces doesn't get a chance to do so anymore. You also change the signal in your handler to get a core dump, look here http://www.alexonlinux.com/how-to-handle-sigsegv-but-also-generate-core-dump
Re: Why sometimes stacktraces are printed and sometimes not?
On Wednesday, 29 September 2021 at 12:15:30 UTC, Steven Schveighoffer wrote: On 9/29/21 6:57 AM, JN wrote: What makes the difference on whether a crash stacktrace gets printed or not? Sometimes I get a nice clean stacktrace with line numbers, sometimes all I get is "segmentation fault error -1265436346" (pseudo example) and I need to run under debugger to get the crash location. segmentation faults are memory access errors. It means you are accessing a memory address that is not valid for your application. If you are accessing the wrong memory, it means something is terribly wrong in your program. Note that on Windows in 32-bit mode, I believe you get a stack trace. On Linux, there is the undocumented `etc.linux.memoryhandler` which allows you to register an error-throwing signal handler. Signals are not really easy to deal with in terms of properly throwing an exception. This only works on Linux, so I don't know if it's possible to port to other OSes. I've also found sometimes that it doesn't work right, so I only enable it when I am debugging. -Steve You might also mention that even if you had a stacktrace where the error happened, that's usually not where the error was caused. It's most likely a completely different place in the code. The only time where you somewhat can be sure where it happens is when you try to access ex. a reference type that hasn't been instantiated. That's why in languages like C# you don't get a segfault/access violation, but you get a NullReferenceException. It's not a concept D has, so it defaults to segfault/access violation. Which means you're in really deep water when you encounter one because you have no idea what caused it and where it was caused.