[Gnu-arch-users] Alternate tagging method, [was arch with 'special files']

[John Meinel]

John Meinel wrote:
> Tom Lord wrote:
>
> >    From: Josh England <address@hidden>
...

> So I would suggest the following changes, have the "categories" source, 
> not-source, source-w, not-source-w
>
> source       is a regex for including a file as source
> not-source   is evalutated after a match, and specifically regects a
>              match
> source-w     uses shell wildcard syntax
> not-source-w should be obvious
>
> Probably better names than source, precious, backup, junk. At least 
> precious should probably be something like "ignored".

Having had some time to think about it in the chair, this is a revised 
category scheme, evaluated in order:

force-ignore
force-disallow
source
disallow
warn
ignore

ignore - say nothing about this file, do not enter the directory, even
         if the file violates 7-bit rules, to arch, this file does not
         exist. (It is a little hard to fit a 7-bit regex to an 8-bit
         file, but I think we can make it work).
disallow - If present, this is a lint failure.
source - The obvious
warn - If present, warn about it's existence, but it is not an error

Then you would want the flags:

unrecognized disallow
untagged-source warn

This lets you handle the case where something doesn't match a regex, and 
do something reasonable with it.
I'm still open to the idea of the "not-" forms, which allow direct 
negation. They would only be evaluated if the normal form succeeded. But 
it starts adding quite a few levels, which may be more confusing than 
necessary. It is straightforward (force- evaluated first, normal, not- 
evaluated after normal).

Also, I would like to see a per-user setting (say 
~/.arch-params/ignored-files). It would have everything *but* source. 
That way I can configure my Mac to always ignore .DS_Store and 
.gdb_history, without having to worry about setting it in *every* 
project config.
I don't think source should be in there, because then if someone checks 
out my tree, the files instantly go missing. But with ignore/disallow a 
checkout is clean. If I use my per-user to allow build output into my 
tree and someone else checks it out and does a build, then there are 
errors. But that is an okay failure mode, it isn't *really* harmful. tla 
only creates source files.

Also, each of these could have a -w form. I'm pretty sure shell 
expansion can be turned into a regex with simple substitution.

Shell => Regex
* => .*
? => .
| => \|
\* => \*
\\ => \\
, => |
+ => \+
[ => \[
] => \]
. => \.
...

Any other regex meta-characters that I've missed can probably just be 
'\' escaped. The , => | is if we want to allow multiple shell expansions 
per line. I certainly don't think we have to. You can always use 
fnmatch, but translating lets the internal code stick to only regexes.

I don't know of any code inside tla that treats junk files differently 
from precious or backup. There was some statement about copying working 
trees, but as far as I know, that isn't done anymore.
If there is a genuine reason to have the different categories, we can 
put them in.
The idea, though, is to give obvious names, and an obvious evaluation 
order. By having the force-* entries, you can override a source regex 
with an ignore one, so that stating exceptions becomes easier.

I think we can leave in the ',.*' and '+.*' as force-ignore.

I think these rules clean up the current tagging method headaches people 
have. The naming scheme is cleaned up a little, and specifically 
ignoring/including files should be pretty easy.

PS> As a minor point, I think all regexes should be full-match, not 
search style. so regex="bc" doesn't match "AbcD". This eliminates the 
need for ^$ everywhere. If necessary they could be added internally.