5 ms·

C++ object model and other internals explained with simple C code

67 points by hanifvirani 16y ago

cawhitworth 16y ago
Worth noting that this is very gcc specific. Implementation details will vary across compilers, but it's not made clear.
- phaedrus 16y ago
  From the introduction, "NOTE: The actual structure of the generated code depends on the computer architecture, operating system and compiler."
faragon 16y ago
Unfortunately, C++ internals change from compiler to compiler, even from same manufacturer between updates, making impossible usable C++ dynamic libraries safely (C calling conventions are standard, while that is not enough for C++, because of class handling). Unless that becomes solved, C++ will never be able to compete with C for library and system development.
eliasmacpherson 16y ago
I will struggle my way through this, having mostly given up on "Inside the C++ Object Model" by Stan Lippman. I have heard it said that Lippman's book is very out of date and a lot of it doesn't apply to modern compilers. Link seems to be down, although some of it is cached by google. Is there a way to use google webcache to edit the hyperlinks in a cached article so that they point to cached versions of the target?
- mahmud 16y ago
  I thought Lippman's book was adequate. And if you're familiar with the implementation of various statically-typed object models (oxymoron?) then you can see a few ways suggest themselves. Depends on what you want to do with this knowledge. If you want to use this knowledge for gains in performance, you will have to get your hands dirty and dig into specific implementations.
  - eliasmacpherson 16y ago
    I guess I'm complaining more about the reader than the writer. It does provide a useful approach to test specific implementations. It gives good insight as to why certain features are to be avoided for performance, ie multiple inheritance, virtual functions under virtual inheritance, RTTI. The section on exception handling says that there are compiler dependent code size increases and speed is impacted (5% - 10%) even if it is not triggered. That contradicts Stroustrup in Design and Evolution of C++ where its stated that it should be the same cost as a function call. I've heard other people say that exception handling is now "free". I should spend less time gaining second hand information and more time trying this stuff out myself.
malkia 16y ago
In our past games, we had extensive C++ serialization (marshalling, pickling, whatever you call it). We had to fight gcc 2.95, metrowerks, and few other compilers putting the "virtual table pointer" at different places - begining of object, end, etc. Then alignment differences (not strictly C++), and bitfields. Hopefully mangling was avoided, as we did not had to reference at runtime the data, we had pointers. It was quite messy, and you could not reuse exported data - had to be for each platform/compiler.
swah 16y ago
I'm getting: PDOException: SQLSTATE[HY000] [2002] Can't connect to local MySQL server through socket '/var/lib/mysql/mysql.sock' (2) in lock_may_be_available() (line 165 of /srv/www/avabodh.com/public_html/includes/lock.inc).
- njs12345 16y ago
  Coral cache: http://www.avabodh.com.nyud.net:8090/cxxin/cxx.html http://www.avabodh.com.nyud.net:8090/cxxin/cxx.html
  - reemrevnivek 16y ago
    That cache just points to the outline for me. How do I get to, say, the part of the article on exception handling? Or, alternatively, the paper on C? Disclaimer: I've never used Drupal or a coral cache before, so I don't know how to create these links. I'm going to try the Google cache...
    - njs12345 16y ago
      To make a coral cache link you just append .nyud.net:8090 to the hostname, as above. Less useful nowadays than it once was, back when a good Slashdotting was guaranteed to cause a server meltdown..
rdtsc 16y ago
It is also worth highlighting that extensive object oriented systems can and have been built on top of C : http://library.gnome.org/devel/gobject/stable/ http://library.gnome.org/devel/gobject/stable/
jpr 16y ago
If C++ can be explained adequately with simple C code, why use C++ in the first place?
- roel_v 16y ago
  C++ cannot be explained fully with simple and safe C code. For example, theoretically one could implement the most basic usage of templates with macros; it would however not be easy to work with (error message in the wrong places) and it would still not emulate e.g. template specialization.
- danssig 16y ago
  Better type safety is possible in C++ than in C.
- mahmud 16y ago
  Some manual C code can adequately explain what C++ does automatically. The difference is cost; you have to link against C code, for some platforms, what C++ provides for free on all platforms.

67 points by hanifvirani 16y ago

Worth noting that this is very gcc specific. Implementation details will vary across compilers, but it's not made clear.

phaedrus 16y ago
From the introduction, "NOTE: The actual structure of the generated code depends on the computer architecture, operating system and compiler."

faragon 16y ago

Unfortunately, C++ internals change from compiler to compiler, even from same manufacturer between updates, making impossible usable C++ dynamic libraries safely (C calling conventions are standard, while that is not enough for C++, because of class handling). Unless that becomes solved, C++ will never be able to compete with C for library and system development.

eliasmacpherson 16y ago

I will struggle my way through this, having mostly given up on "Inside the C++ Object Model" by Stan Lippman. I have heard it said that Lippman's book is very out of date and a lot of it doesn't apply to modern compilers. Link seems to be down, although some of it is cached by google. Is there a way to use google webcache to edit the hyperlinks in a cached article so that they point to cached versions of the target?

mahmud 16y ago
I thought Lippman's book was adequate. And if you're familiar with the implementation of various statically-typed object models (oxymoron?) then you can see a few ways suggest themselves. Depends on what you want to do with this knowledge. If you want to use this knowledge for gains in performance, you will have to get your hands dirty and dig into specific implementations.
- eliasmacpherson 16y ago
  I guess I'm complaining more about the reader than the writer. It does provide a useful approach to test specific implementations. It gives good insight as to why certain features are to be avoided for performance, ie multiple inheritance, virtual functions under virtual inheritance, RTTI. The section on exception handling says that there are compiler dependent code size increases and speed is impacted (5% - 10%) even if it is not triggered. That contradicts Stroustrup in Design and Evolution of C++ where its stated that it should be the same cost as a function call. I've heard other people say that exception handling is now "free". I should spend less time gaining second hand information and more time trying this stuff out myself.

malkia 16y ago

In our past games, we had extensive C++ serialization (marshalling, pickling, whatever you call it). We had to fight gcc 2.95, metrowerks, and few other compilers putting the "virtual table pointer" at different places - begining of object, end, etc. Then alignment differences (not strictly C++), and bitfields. Hopefully mangling was avoided, as we did not had to reference at runtime the data, we had pointers. It was quite messy, and you could not reuse exported data - had to be for each platform/compiler.

swah 16y ago

I'm getting: PDOException: SQLSTATE[HY000] [2002] Can't connect to local MySQL server through socket '/var/lib/mysql/mysql.sock' (2) in lock_may_be_available() (line 165 of /srv/www/avabodh.com/public_html/includes/lock.inc).

njs12345 16y ago
Coral cache: http://www.avabodh.com.nyud.net:8090/cxxin/cxx.html http://www.avabodh.com.nyud.net:8090/cxxin/cxx.html
- reemrevnivek 16y ago
  That cache just points to the outline for me. How do I get to, say, the part of the article on exception handling? Or, alternatively, the paper on C? Disclaimer: I've never used Drupal or a coral cache before, so I don't know how to create these links. I'm going to try the Google cache...
  - njs12345 16y ago
    To make a coral cache link you just append .nyud.net:8090 to the hostname, as above. Less useful nowadays than it once was, back when a good Slashdotting was guaranteed to cause a server meltdown..

rdtsc 16y ago

It is also worth highlighting that extensive object oriented systems can and have been built on top of C : http://library.gnome.org/devel/gobject/stable/ http://library.gnome.org/devel/gobject/stable/

jpr 16y ago

If C++ can be explained adequately with simple C code, why use C++ in the first place?

roel_v 16y ago
C++ cannot be explained fully with simple and safe C code. For example, theoretically one could implement the most basic usage of templates with macros; it would however not be easy to work with (error message in the wrong places) and it would still not emulate e.g. template specialization.
danssig 16y ago
Better type safety is possible in C++ than in C.
mahmud 16y ago
Some manual C code can adequately explain what C++ does automatically. The difference is cost; you have to link against C code, for some platforms, what C++ provides for free on all platforms.