Generator output contains duplicate lines #205

Activity

1. huush commented on Nov 20, 2023

   

   huush

   on Nov 20, 2023

   Author

   More actions

   Also, the linked Generator appears to not pick up the files in Global and Community

   👍React with 👍1Reacted by Alexander Karatarakis
2. alexkaratarakis commented on Dec 3, 2023

   

   alexkaratarakis

   on Dec 3, 2023

   Last edited by alexkaratarakis

   Member

   More actions

   What happens if there are then multiple lines of "* text=auto" throughout the file? Does this cause some form of reset of the overrides? Can this generated file actually be used, or do we first need to filter this combined file and remove all duplication?

   That is a good point! Let's explore. I made a toy repository with just 2 files:

   .gitattributes
    test.cpp
   

   git check-attr --all test.cpp returns what git treats the file as. Here are the results for a given .gitattribute file:
   (git version 2.34.1)

   (1)

   * text=auto
   *.cpp binary
   *.cpp text
   

   test.cpp: binary: set
   test.cpp: diff: unset
   test.cpp: merge: unset
   test.cpp: text: set
   

   (2)

   *.cpp binary
   *.cpp text
   * text=auto
   

   test.cpp: binary: set
   test.cpp: diff: unset
   test.cpp: merge: unset
   test.cpp: text: auto
   

   (3)

   * text=auto
   *.cpp text
   *.cpp binary
   

   test.cpp: binary: set
   test.cpp: diff: unset
   test.cpp: merge: unset
   test.cpp: text: unset
   

   (4)

   *.cpp text
   *.cpp binary
   * text=auto
   

   test.cpp: binary: set
   test.cpp: diff: unset
   test.cpp: merge: unset
   test.cpp: text: auto
   

   The results are interesting. It seems like the last entry takes precedence, so text is either auto or set depending on the last entry. But also binary seems to "win" regardless of where it is. binary is always set in the above cases and it even unsets text when it comes last.

   This doesn't look immediately obvious, so a Generator could flatten and deduplicate entries to avoid possible confusion - and have the entry match what git would have done if all entries where present.
   On the other hand, having separate sections for each gitattribute that you used is very similar to what one would do manually (without a generator) and allows further manual modification and future update because any section would be exactly as you found it from the source.
   I think it could go either way (maybe an option?).

   Note that the behavior would be easily observable. If binary is set, then we get the following with git diff:

   diff --git a/test.cpp b/test.cpp
   index cc3f56d..1df2af6 100644
   Binary files a/test.cpp and b/test.cpp differ
   

   whereas if it is not set, then:

   diff --git a/test.cpp b/test.cpp
   index cc3f56d..1df2af6 100644
   --- a/test.cpp
   +++ b/test.cpp
   @@ -1 +1,2 @@
    // blah
   +// blah2
   

   Similarly, with customized diff options, the diff would show the "winning" setting.
3. alexkaratarakis commented on Dec 3, 2023

   

   alexkaratarakis

   on Dec 3, 2023

   Last edited by alexkaratarakis

   Member

   More actions

   Also, the linked Generator appears to not pick up the files in Global and Community

   You are right. The subfolders are recent additions. The generator(s) have not been updated to take this into account.