implement patch manually without using external software #136
Conversation
| let original = | ||
| fs::read_to_string(out_dir.join(original)).expect("Unable to read original file"); | ||
| let original = dos2unix(&original); | ||
| match apply(&original, &patch) { |
There was a problem hiding this comment.
Does it support fuzzy patch applying?
Could you test build with parallel feature on windows?
There was a problem hiding this comment.
Not sure about fuzzy patching, but I am building it with parallel and futures
There was a problem hiding this comment.
The fact is that MSVC patches changes line numbers and may interfere applying other patches.
| diffy = "0.3.0" | ||
| newline-converter = "0.3.0" |
There was a problem hiding this comment.
Avoid patch number in dependency version.
sys/build.rs
Outdated
| .expect("Cannot find modified file name"); | ||
|
|
||
| // for now | ||
| assert_eq!(original, modified); |
| stdin.write_all(&patch).expect("Unable to apply patch"); | ||
| let out_dir = out_dir.as_ref(); | ||
| let patch = fs::read_to_string(patch).expect("Unable to read patch"); | ||
| let patch = dos2unix(&patch); |
There was a problem hiding this comment.
Does we actually need that for both patch and source?
Does we need it at all?
I mean we may simply fix our patches if you have problem with it.
There was a problem hiding this comment.
How do you determine which file to patch?
There was a problem hiding this comment.
@stevefan1999-personal Course, I about line-ending conversion.
There was a problem hiding this comment.
@stevefan1999-personal Course, I about line-ending conversion.
I'm not sure what you mean
There was a problem hiding this comment.
Why we need the following:
let patch = dos2unix(&patch);and
let original = dos2unix(&original);There was a problem hiding this comment.
I have CRLF in Windows and diffy does not support CRLF yet. There is no allocation cost if you have LF.
There was a problem hiding this comment.
I know this could turn political quickly, but since QuickJS is written with GCC in mind, and is written by Fabrice Bellard, a legendary open source developer, I'm pretty sure he wrote this while only running on Linux, and Linux is LF by default.
Yet on Git and Windows, you have automatic LR to CRLF conversion. It is up to the user to turn it off, but it seems like it is a viral option, and I believe this would cause even more unwanted trouble to end-users.
I'm not aware if autocrlf can be turned off per Git repository or can be disabled for Git submodules but anyway, it is almost cost-free if you run this on Unix-like operating system. We should use do this conversion to "normalize" anyway.
There was a problem hiding this comment.
Using this .gitattributes will, by default, force text files to use LF EOLs, with CRLF exceptions where required for WinOS or MSVC.
# .gitattributes (git attributes config file)
# default; use LF EOLs for text files
* text=auto eol=lf
# CRLF required; force required CRLF EOLs for WinOS BAT/CMD and MSVC SLN/VCPROJ/VCXPROJ files
*.[Bb][Aa][Tt] text eol=crlf
*.[Cc][Mm][Dd] text eol=crlf
*.[Ss][Ll][Nn] text eol=crlf
*.[Vv][Cc][Pp][Rr][Oo][Jj] text eol=crlf
*.[Vv][Cc][Pp][Rr][Oo][Jj].* text eol=crlf
*.[Vv][Cc][Xx][Pp][Rr][Oo][Jj] text eol=crlf
*.[Vv][Cc][Xx][Pp][Rr][Oo][Jj].* text eol=crlf
This PR attempts to manually do patches without using GNU Patch. Despite the algorithm is quite primitive and had some questionable assumptions, at least it does compile in Windows natively now. The edge case is that I don't know how to determine the correct file to patch -- so I just assumed the
a/{filename}andb/{filename}format.Fixes #88. Since I made this in Windows and I have to compile it by myself. So I can't possibly made this PR possible if there are any errors. No need to test for Windows therefore.