01:17 <ltr_> hello

08:33 <pupp> ltr_, hello

10:22 <J> Hello

11:07 <tankfeeder> hi all

17:22 <dimenus> andrewrk: ran into a few snags, not sure how I had this working properly the other day but it doesn't work now.

17:22 <dimenus> I can compile a program linking to the CRT just fine, but it doesn't find symbols as I would expect

21:31 <GitHub80> [zig] Dimenus opened pull request #570: WIN32: Linking with the CRT (master...master) https://git.io/vFqwC

21:43 <andrewrk> dimenus, I'm looking at your pull request now

21:44 <dimenus> I think it closes a couple other issues as well

21:44 <dimenus> but i haven't hunted around yet

21:44 <dimenus> i figured out the printf issue by the way, silly Microsoft

21:46 <andrewrk> printf issue, sounds like text/binary mode

21:48 <dimenus> i was getting undefined reference to printf and it was driving me nuts

21:49 <dimenus> created a hello world C app and removed the '%AdditioanlDependencies' included by default in Visual Studio

21:49 <dimenus> and that too failed the same way

21:49 <dimenus> so I found out that they ported the implementation of non-inline stdio to another lib

21:49 <dimenus> so really we're linking 4 (5 if you include the kernel32 that zig builds) libs in order to support the CRT

21:50 <andrewrk> ah, good to know.

21:52 <andrewrk> dimenus, I think that means once we solve https://github.com/zig-lang/zig/issues/515 we could avoid linking that lib that has printf

21:56 <dimenus> correct

22:40 <PV_> andrewrk: I have a question about comptime string paramater, like fn(comptime par : []const u8). Is such parameter always string sitting in readonly section of executable?

22:40 <PV_> Or is there chance that compiler will put it somehow on stack? What I ask is whether it is safe to take address of such string parameter and use it outside the function.

22:41 <andrewrk> PV_, yes if it is comptime it is guaranteed to be in static readonly memory

22:42 <andrewrk> it seems like a smell to do that though, maybe you can show some code and I can have a look at what you're trying to do

22:43 <PV_> OK. What I intend to do is logger which takes a string and instead of copying it it stores only the address. I wasn't sure whether it is safe enough.

22:54 <andrewrk> PV_, and then what? what does it do with the address?

22:56 <PV_> My aim is to implement white box testing support. Functions log various things, the test then checks whether what was expected to happen did really happen. The logs are stored as pointers (where applicable) to save space.

22:59 <andrewrk> interesting

23:04 <PV_> I got this idea from Agaram Kertik ( http://akkartik.name/post/tracing-tests ). Basically it allows to verify what's going on inside the code, without need to change the code, make everything exposed, everything interface. One example: If allocators do log their allocations and dellocations then it is possible to verify that given code snippet does not use heap or uses it exactly as planned.

23:05 <andrewrk> that's a neat idea

23:10 <PV_> I'd implemented it in C and it is performant enough. In a test I first express interest in certain logs (in certain string prefixes), the code then records these logs (and not the others) and then I check if the expected logs are in and undesirable logs are not present. These limitations keep the overhead down.

23:11 <PV_> The essay I linked describes more powerful system (everything is logged, one can browse the logs to get idea what is going on). K. Agaram later wrote me that he switched to restricted system, like I have.

23:14 <PV_> The technique is non-intrusive, no need to modify anything, only the logs need to be added. It is prone to typos, however. I have some hope for metaprogramming in Zig to be able to check the strings and spot obvious mistakes.

23:30 <andrewrk> PV_, you should be able to do it with 0 runtime cost, you can check for builtin.is_test

23:31 <andrewrk> or make your own comptime known bool

23:44 <PV_> andrewrk: my logs are generated only inside tests, and also only if the were requested (their string prefix was requested). I played with idea to place assembler "nop" instructions instead all logs and then re-insert the original log calls according to the need, but gave up.