CR-LF in the codebase

[neheb]

Is there a good reason for using CRLF for the source? I ask since CRLF can cause issues with git.

[metalefty]

+1 to LF, I prefer LF.

[chipitsine]

I was sure .gitattributes keep us safe. Any example of how it could go wrong?

[neheb]

For OpenWrt, we have separate patch files that get applied to the main package.

Using git diff > .patch works with git config core.autocrlf true

When it comes time to commit to the repository though, core.autocrlf must be enabled as well. Otherwise git will change the line endings to LF. No amount of configure changes and git commit --amend will help after that.

[chipitsine]

that option does not help ?

https://github.com/SoftEtherVPN/SoftEtherVPN/blob/master/.gitattributes

btw, what are those patches ? can you provide a link ?

[chipitsine]

@neheb , do you mean https://github.com/neheb/packages/tree/master/net/softethervpn5 ?

[neheb]

No. https://github.com/neheb/packages/tree/soft5/net/softethervpn5/patches

Last two

edit: note that a slightly different solution was merged.

[chipitsine]

so, I have several questions

1. the last two looks like they might be included in upstream, @davidebeatrici ?
2. if you are talking about CR-LF, you mean only limited number of files ? do you mind to open a PR ?
3. we definetly will appreciate some help to setup openwrt build in our travis-ci matrix (or maybe openwrt uses some other kind of CI)

[neheb]

Well, I'm only asking why CRLF is used in the codebase at all. My guess is for Windows development. I would guess that with Windows 10, Visual Studio supports LF just fine, given that Notepad does.

Travis used to be used. But it had hard to debug build failures (was using an old Ubuntu LTS). OpenWrt uses CircleCI. The configuration file is in the root directory.

Since I got this to build, I noticed it does not compile with OpenSSL lacking deprecated APIs. I might whip up a patch to fix.

edit: Also note that while the first two patches are correct, the last two have mangled end-line characters and thus do not apply. Locally, this is incorrect but they got mangled during git commit.

[Andy2244]

A year ago or so this issue was also noted by me and while the .gitattributes was changed, the conclusion was that "someone" would have to recommit the whole project with LF endings, which never happened.

So yes while you create diffs/patches, the result has mixed endings. This works fine unless some form of auto conversion happens either via texteditor or on commit. So you need to take extra care not to modify the patches.

PS: The first two patches we use are to somewhat mitigate #575.

[neheb]

I figured out that vi does not mess with the line endings. Everything else does.

[chipitsine]

if you feel better about line ending conversion, please provide such patch

[neheb]

Meh this is a minor point. It's only really relevant to the OpenWrt workflow where GNU patch and git differ. Closing.