jcpp
jcpp copied to clipboard
shevek
Various fixes and additions
Added code to find and include files from frameworks Added code to support the #import directive and only process files once when using that directive Added code to handle ?: condition expression evaluation Fixed a bug when macro arguments spanned over multiple lines Fixed a bug where SQSTRING's was not passed through Fixed a bug where something that started as a number could span an entire line eating up unrelated tokens
The lexing to end-of-line on error was relatively deliberate. I'm not sure that returning IDENTIFIER is any better. Perhaps the INVALID should only reach until the next whitespace?
shevek@shadow:~/java/jcpp$ git ci -m "Preprocessor: Fix SQSTRING." [master 9c8391a] Preprocessor: Fix SQSTRING. 2 files changed, 16 insertions(+), 6 deletions(-)
Re: Frameworks, all the file I/O including the pre-checks for existence have to be done through VirtualFileSystem - breaking out and using java.io.File will break the VFS abstraction.
"Fixed a bug when macro arguments spanned over multiple lines" - I'm not seeing this... where should I look?
[master 2db1eaf] LexerSource: Handle invalid number as a single INVALID token and don't consume the entire line. 3 files changed, 44 insertions(+), 15 deletions(-)
Handled using unicode rules rather than simple C rules.
As to the errornous number suffix i think there was some version information in some of Apples iOS headers that was in a form similiar to 5_1_0 and only used by the preprocessor, since jcpp took the entire line when faulting on the suffix this broke things. (It was a few months ago i was working on this, i'll try taking in your code later today and test if it still parses through everything)
Again, not remembering everything exactly but i think that it was without the "case NL" around line 743 in commit(d67cee2) that macro arguments spanning over multiple lines broke since the preprocessor would back-off when seeing something "unknown".
I think there was something that made working with the VFS hard in this case and i think i reasoned that the abstraction would be to allow compile servers that abstracted parsing user code as f.ex. coming from uploaded zip's but still used local "system" packages. I'll take a peek later to see if i can figure out what made it tricky to use the VFS for this.
(As the INVALID vs IDENTIFIER thing, i think i initially changed to IDENTIFIER when i didn't notice what was wrong but changed it back to invalid in a4ce2c8 )
[master cccd60f] Handle conditionals in preprocessor statements. 2 files changed, 16 insertions(+), 3 deletions(-)
I think that given #20 and a re-look at #import, I think that will cover everything here?
What happens if I hash #import on VirtualFile instead of String?
Can you #import <> and then #import "" the same thing? The implementation in this PR implies 'yes', but that's not quite logical.
The code for the frameworks seemingly looks fine to me, simpler than my variant also that was overly defensive and kinda went in the wrong way (I think my variant was inspired by one of the other forks)
hashing on virtualfile sounds more sane but i remember some weirdness described somewhere (as to why i separated the import variants, i'll try to find the source for this).
Actually, i think that you're right about just going with the VFS identity. I separated <> and "" because i didn't think i could rely on simple names (and didn't know that VFS entries were supposed to be reliable for identity checking) in the case someone included a local copy after a system copy and so i erred on the side of extra inclusions.
OK, so I'll put the framework code into master, please let me know whether it works as I have no objc code to play with.
I'll also code up import to use VFS identity (and perhaps to share the cache with #pragma once).
When I've done the #import work, I'd love your opinion on whether there's anything in your commit that I missed?