Fixing OutputFile comment: path is relative to working directory. #26
Conversation
googlebot
added
the
cla: yes
|
This would actually be a spec change. I don't think we can chalk this up as a comment fix. |
|
It's not a spec change. You can see it being documented correctly in line https://github.com/bazelbuild/remote-apis/blob/master/build/bazel/remote/execution/v2/remote_execution.proto#L431, and the current line just straight-up contradicts that one, because I forgot to change it. |
|
I clearly overlooked something about the proposal on Paths + Whole directory special case. I assumed wrongly that absolute pats would refer to the input_root. |
|
But working directory is equal to the input root by default? This is more
related to the #8 change.
…On Thu, Aug 23, 2018 at 5:33 AM Sander Striker ***@***.***> wrote:
I clearly overlooked something about the proposal on Paths + Whole
directory special case
<https://docs.google.com/document/d/12c3oAPgedckLpue2yj0xGgJTEOyUm4mXWWBJ157J-8I/edit#heading=h.sg3la2izb6yt>.
I assumed wrongly that absolute pats would refer to the input_root.
For the BuildStream client we have a use case where we do want to retrieve
files relative to the input_root. Shall I file a new issue and proposal for
that?
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#26 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AYoKuNoEBXKv1DrS9IzuyIhszzqsjdcnks5uTncBgaJpZM4WIEEL>
.
|
|
@ola-rozenfeld that sounds about right. Now, to refer to the input_root for output_*, should I open up a new proposal that allows passing of absolute paths that are rooted in input_root? Or should I pass in ../../../subdir_of_input_root for output_directory? The latter is legit I think according to spec, but a bit awkward. |
|
You are talking about the case where the working directory is a subdir of input root, right? For that one, you pass a ../other_subdir_of_input_root/myoutput path. I think when we discussed "Paths + Whole directory special case" we decided that using absolute paths to refer to something else, like input root, was too confusing. |
…azelbuild#26) * A Chunker to support memory-bounded buffered uploads with lazy file reads. * Addressing offline comments * Addressing comments * Making the Digest public instead of just the Size * Reconsidered, I don't need to have Digest as part of Chunk. * Adding the post-Next-reliability contraint into the unit tests. Enabling more tests. * Stupid Go won't print an int into a %s...
Fixes #25.