digitalmars.D.learn - How to make a slow compile
- monkyyy (9/9) Aug 20 once again dconf contains people who talk about compile speed as
- Kapendev (3/12) Aug 20 No idea. Maybe they should use shell + awk + sed more. This is
- Brother Bill (3/5) Aug 20 Compiling Hello world in D is a little bit slower that Go
- monkyyy (9/14) Aug 20 Thats the runtime and std self importing everything and if your
- Kapendev (2/7) Aug 20 C with tcc is faster than D an GO.
- evilrat (12/21) Aug 20 I have godot-dlang builds (with dub/LDC) compiling for about 3
- Monkyyy (6/31) Aug 21 Can you post what you believe is slow?
- Andrey Zherikov (4/13) Aug 24 My recent experience: removing usage of SumType and replacing it
- monkyyy (3/19) Aug 24 What sort of difference did you do? Did the memory usage matter
once again dconf contains people who talk about compile speed as if it can be slow What are they doing? Even my worse compile time abstraction was O(n^2) or maybe some awful string concatenation of an entire file; still effectively instant Wheres a 1000 line file that does something meta programmingly useful that takes 5 minutes? Do they lose track of what the compiler is even doing? Do they turn off the template recursion depth limit?
Aug 20
Kapendev <alexandroskapretsos gmail.com> On Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:once again dconf contains people who talk about compile speed as if it can be slow What are they doing? Even my worse compile time abstraction was O(n^2) or maybe some awful string concatenation of an entire file; still effectively instant Wheres a 1000 line file that does something meta programmingly useful that takes 5 minutes? Do they lose track of what the compiler is even doing? Do they turn off the template recursion depth limit?No idea. Maybe they should use shell + awk + sed more. This is how chads generate code **blazingly** fast.
Aug 20
On Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:once again dconf contains people who talk about compile speed as if it can be slowCompiling Hello world in D is a little bit slower that Go language.
Aug 20
On Wednesday, 20 August 2025 at 16:57:18 UTC, Brother Bill wrote:On Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:Thats the runtime and std self importing everything and if your being wrong dub taking an entire 1/2 second to consider how to to call one of the easiest to work with compilers Th dub thing is straight stupid, the std should have a dag of acceptable imports, and the runtime is unfortunate fixed cost. But even if dub is in play with its unacceptable half sec to read a 10 line json, it isnt the "90 minutes" weka compile they talk about.once again dconf contains people who talk about compile speed as if it can be slowCompiling Hello world in D is a little bit slower that Go language.
Aug 20
On Wednesday, 20 August 2025 at 16:57:18 UTC, Brother Bill wrote:On Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:C with tcc is faster than D an GO.once again dconf contains people who talk about compile speed as if it can be slowCompiling Hello world in D is a little bit slower that Go language.
Aug 20
On Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:once again dconf contains people who talk about compile speed as if it can be slow What are they doing? Even my worse compile time abstraction was O(n^2) or maybe some awful string concatenation of an entire file; still effectively instant Wheres a 1000 line file that does something meta programmingly useful that takes 5 minutes? Do they lose track of what the compiler is even doing? Do they turn off the template recursion depth limit?I have godot-dlang builds (with dub/LDC) compiling for about 3 minutes on top hardware. But this is more of a config issues and how LDC handles all that code symbols(functions, etc...), having to build a spaghetti templates for over 1k classes is no joke. This is something i am planning to look into, having a tiny WASM game scripts for 30 Mb shared library is no fun. This is mainly because it uses dub import-paths AND source-paths together for usability simplicity reasons. Would be much faster to remove source-paths and properly configure D's -i include pattern, that requires figuring dub configuration that properly uses predefined env variables.
Aug 20
On Thursday, 21 August 2025 at 05:38:08 UTC, evilrat wrote:On Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:once again dconf contains people who talk about compile speed as if it can be slow What are they doing? Even my worse compile time abstraction was O(n^2) or maybe some awful string concatenation of an entire file; still effectively instant Wheres a 1000 line file that does something meta programmingly useful that takes 5 minutes? Do they lose track of what the compiler is even doing? Do they turn off the template recursion depth limit?I have godot-dlang builds (with dub/LDC) compiling for about 3 minutes on top hardware. But this is more of a config issues and how LDC handles all that code symbols(functions, etc...), having to build a spaghetti templates for over 1k classes is no joke. This is something i am planning to look into, having a tiny WASM game scripts for 30 Mb shared library is no fun. This is mainly because it uses dub import-paths AND source-paths together for usability simplicity reasons. Would be much faster to remove source-paths and properly configure D's -i include pattern, that requires figuring dub configuration that properly uses predefined env variables.spaghetti templateCan you post what you believe is slow? Can dub cause 3 minute compiles, I would've thought it would O() based on file count (reading json badily into concat a unnesserily complex compile commmand) is dub some how worse then I thought possible?
Aug 21
On Thursday, 21 August 2025 at 13:52:32 UTC, Monkyyy wrote:On Thursday, 21 August 2025 at 05:38:08 UTC, evilrat wrote:Like i've said it is from the fact that the project tries to build over 1k classes, an entire godot API. It does templated binding calls, parameter conversions, class registration, etc... This is also related to LDC's way of managing symbol visibility for shared libraries, it simply does not knows what needed, except maybe the simplest cases, can be partially solved with doing `dub --build-mode=allAtOnce`. DMD on the other hand does all this in just seconds. here is a cold full rebuild using DMD on windows `godot-dlang> measure-command { dub build :test }` `TotalSeconds : 13,9210259` and a build after minor change `TotalSeconds : 1,7115313`On Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:once again dconf contains people who talk about compile speed as if it can be slow What are they doing? Even my worse compile time abstraction was O(n^2) or maybe some awful string concatenation of an entire file; still effectively instant Wheres a 1000 line file that does something meta programmingly useful that takes 5 minutes? Do they lose track of what the compiler is even doing? Do they turn off the template recursion depth limit?I have godot-dlang builds (with dub/LDC) compiling for about 3 minutes on top hardware. But this is more of a config issues and how LDC handles all that code symbols(functions, etc...), having to build a spaghetti templates for over 1k classes is no joke. This is something i am planning to look into, having a tiny WASM game scripts for 30 Mb shared library is no fun. This is mainly because it uses dub import-paths AND source-paths together for usability simplicity reasons. Would be much faster to remove source-paths and properly configure D's -i include pattern, that requires figuring dub configuration that properly uses predefined env variables.spaghetti templateCan you post what you believe is slow? Can dub cause 3 minute compiles, I would've thought it would O() based on file count (reading json badily into concat a unnesserily complex compile commmand) is dub some how worse then I thought possible?
Aug 21
On Thursday, 21 August 2025 at 13:52:32 UTC, Monkyyy wrote:On Thursday, 21 August 2025 at 05:38:08 UTC, evilrat wrote:https://github.com/Circular-Studios/Dash here is a good example, this project does heavily rely on reflection, similar to godot-dlang but going even further to also do serialization and other things, all that resulted in long builds. though it'll probably no longer compiles without major rewrite, i remember it was like 40 seconds builds after every source change. https://github.com/Circular-Studios/DashOn Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:once again dconf contains people who talk about compile speed as if it can be slow What are they doing? Even my worse compile time abstraction was O(n^2) or maybe some awful string concatenation of an entire file; still effectively instant Wheres a 1000 line file that does something meta programmingly useful that takes 5 minutes? Do they lose track of what the compiler is even doing? Do they turn off the template recursion depth limit?I have godot-dlang builds (with dub/LDC) compiling for about 3 minutes on top hardware. But this is more of a config issues and how LDC handles all that code symbols(functions, etc...), having to build a spaghetti templates for over 1k classes is no joke. This is something i am planning to look into, having a tiny WASM game scripts for 30 Mb shared library is no fun. This is mainly because it uses dub import-paths AND source-paths together for usability simplicity reasons. Would be much faster to remove source-paths and properly configure D's -i include pattern, that requires figuring dub configuration that properly uses predefined env variables.spaghetti templateCan you post what you believe is slow?
Aug 21
On Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:once again dconf contains people who talk about compile speed as if it can be slow What are they doing? Even my worse compile time abstraction was O(n^2) or maybe some awful string concatenation of an entire file; still effectively instant Wheres a 1000 line file that does something meta programmingly useful that takes 5 minutes? Do they lose track of what the compiler is even doing? Do they turn off the template recursion depth limit?My recent experience: removing usage of SumType and replacing it with my own solution reduced dmd memory usage from 6GB to 1.7GB and improved compilation speed.
Aug 24
On Sunday, 24 August 2025 at 13:59:46 UTC, Andrey Zherikov wrote:On Wednesday, 20 August 2025 at 15:33:45 UTC, monkyyy wrote:What sort of difference did you do? Did the memory usage matter for speed at all?once again dconf contains people who talk about compile speed as if it can be slow What are they doing? Even my worse compile time abstraction was O(n^2) or maybe some awful string concatenation of an entire file; still effectively instant Wheres a 1000 line file that does something meta programmingly useful that takes 5 minutes? Do they lose track of what the compiler is even doing? Do they turn off the template recursion depth limit?My recent experience: removing usage of SumType and replacing it with my own solution reduced dmd memory usage from 6GB to 1.7GB and improved compilation speed.
Aug 24