#tyndur · 2009-12-09

88 Nachrichten · Seite 1 von 1 · Original-CSV herunterladen

  1. pcmattmankwolf: Out of interest, when do you expect to have English documentation available for CDI?
  2. kwolfWhen quok is done ;-)
  3. pcmattmanIs he doing translation, or are you waiting? :)
  4. kwolfSeriously, I'm not sure. I don't want to translate comments for things that are going to change anyway
  5. kwolfHe said he'd start to translate some of them
  6. pcmattmanTrue
  7. kwolfWhile he's implementing CDI in his own OS
  8. pcmattmanIs there anything important I missed while I was away?
  9. pcmattmanYes, he mentioned that today
  10. kwolfeddyb somehow stopped to have his monologues and we couldn't figure how to get them back ;-)
  11. kwolfOtherwise not a lot of important things
  12. pcmattmanOkay
  13. kwolfOh, except for the driver initialisation patches, of course
  14. pcmattmanYes, I noticed them in the list
  15. kwolfCDI_DRIVER instead of pedigree.cc
  16. kwolfHope it works out for you
  17. pcmattmanMuch nicer! :)
  18. pcmattmanI might have a look at getting that integrated now, actually
  19. pcmattmankwolf: ping
  20. kwolfpong
  21. pcmattmanJust hit a bit of a problem in include/cdi/dma.h
  22. kwolfUnrelated to the recent changes then?
  23. pcmattmanYour definition of cdi_dma_handle contains a "FILE*" member, which is not defined
  24. pcmattmanYes, unrelated
  25. pcmattmanAt this stage, I am merely going to make an empty struct in Pedigree's cdi-osdep.h, but I thought I would bring it up anyway
  26. kwolfRight, this belongs into cdi-osdep.h
  27. kwolfThere even is a comment "LOST-Implementation..."
  28. pcmattmanYup
  29. kwolfWhich is (a) the wrong name of the OS and (b) not correct German, because it's correctly called "Implementierung" ;-)
  30. pcmattmanHehe :)
  31. pcmattmanIt seems Pedigree also requires specific elements in the structure of dma.h
  32. pcmattmanI am preparing a set of patches as a result of my recent changes, which should make things a bit neater
  33. kwolfThanks. I think we really need more cleanup patches in the next time
  34. pcmattmanWell, re-integrating the latest git CDI into Pedigree shows that kind of thing really well
  35. kwolfHm... I'm wondering if we need to define struct cdi_dma_handle publicly at all
  36. kwolfThe interfaces look like it could be completely opaque
  37. pcmattmanPossibly
  38. kwolfThe next step I'm going to make in restructuring is to make device creation library driven.
  39. pcmattmanWhat do you mean by that
  40. pcmattman?
  41. kwolfThat is, instead of each driver searching its devices in its init function, I'd prefer to have the CDI library call a probe function
  42. pcmattmanOh, I see
  43. kwolfWhich gets the PCI device as a parameter
  44. kwolfThis will help with hotplugging, which we definitely need as we go forward with USB
  45. pcmattmanYup
  46. kwolfMaybe this weekend, don't know. Probably depends on how things are going with our 0.2.2 release
  47. pcmattmanHmm, kwolf: There's a mismatch between the documentation of CDI_DRIVER and the actual calls in drivers
  48. kwolfHm
  49. kwolf?
  50. pcmattmanThe documentation says "Pointer to a driver description (struct cdi_driver*)" but it's called with a non-pointer object
  51. kwolfOops. Care to send a patch?
  52. kwolfThe question is which side to fix. I tend to fixing the comment
  53. pcmattmanI'll patch the comment
  54. pcmattmanHowever, I have run into a much bigger problem now
  55. pcmattmanAt the moment the cdi_*_driver struct is built as a static object, where all of the relocations for init/destroy/init_device/remove_device will happen when the module is loaded
  56. pcmattmanHowever, because of the way our modules work, we need a const initialiser for each component, which means we can't use a struct
  57. kwolfWhat you need is a init function, right?
  58. pcmattmanThat already technically exists, it just can't be specified in a struct
  59. kwolfYour CDI_DRIVER macro would become ugly, but it could define a wrapper function that accesses the struct
  60. pcmattmanThat would be a horrendously ugly macro
  61. pcmattmanI'm working on a possible solution that I'll add as a patch, you can decide if it's okay to keep or not
  62. kwolfWould it really be horrendously ugly?
  63. kwolfNot just the usual ugliness with preprocessor magic?
  64. pcmattmanIn comparison, yes
  65. pcmattmanAnd I doubt Pedigree is the only OS that will run into such a problem
  66. kwolfI would have preferred not adding to many (unnamed) parameters to CDI_DRIVER. Not sure if you can find a solution that avoids this
  67. pcmattmanI have one potential solution, let's see how it goes after I mail the patch to the list
  68. pcmattmanOkay, the mail is out
  69. pcmattmanAnd I'm off to sleep, night
  70. kwolfNicht
  71. kwolf*Night
  72. kwolf_Hi quok :-)
  73. quokHi :)
  74. FreakyPenguinHi
  75. XanClicHi
  76. kwolf_quok: Have you seen Matt's mail, especially his CDI_DRIVER_CONST proposal?
  77. kwolf_Actually, not only quok, but everyone here
  78. XanClicSeen: yes; understood: no ;-)
  79. quokI'm with XanClic on that. I've seen the mail and the diffs, don't really understand the issue though
  80. kwolf_His problem is that he needs to have something like MODULE_INIT(init_function) for Pedigree
  81. kwolf_And he can't read the init_function from the driver struct at compile time
  82. kwolf_I suggested that CDI_DRIVER could generate a static wrapper function that reads it from the struct, but he wasn't quite happy with that
  83. XanClicSo this CONST macro is just the same as the other macro but you give the content of the driver structure instead of the structure itself?
  84. kwolf_Right, I think this is what he's intending
  85. quokCrap, I forgot I have a doctor appointment. I'll bbl. :)
  86. kwolf_Ok
  87. DerHartmutGood evening
  88. XanClicGood evening, DerHartmut