Hi George,
Op di 18 jul 2023 om 20:20 schreef George Gensure <
wer...@gmail.com>:
> A user reported to me that an AOSP RBE invocation fails a validation step that I added in buildfarm to enforce `[working_directory] must be a directory which exists in the input tree.` The working_directory in question was given as '.', which has no special treatment in the wording of the path segments of the identifier, neither does the presumed '..' that will inevitably follow.
>
> Should we treat '.' and '..' as special directory names in working_directory specification? Clearly some clients have taken this liberty already.
Even though Buildbarn supports '.' and '..' in 'working_directory', I
would have loved it if we could have restricted the specification to
say that 'working_directory' can only be empty, or a '/' delimited
string of regular filenames (so not '.' or '..'). But my fear is that
that ship has sailed. So if we want we can extend the spec to clearly
allow its use.
Maybe for REv3 we should make life easier for ourselves by simply
using a repeated string field, where each field has to be a regular
filename. This removes the ambiguity entirely, while removing the
assumption that forward slashes are used. So as follows:
repeated string working_directory = 6;
--
Ed Schouten <
e...@nuxi.nl>