On Wed, Nov 30, 2016 at 4:08 AM, Bernhard Nortmann bernhard.nortmann@web.de wrote:
Hi Joe!
Thanks for chiming in, especially seeing that you have previously worked on something very similar.
Sure! Sorry I didn't review this sooner.
Am 27.11.2016 um 19:53 schrieb Joe Hershberger:
On Wed, Nov 16, 2016 at 4:29 AM, Bernhard Nortmann bernhard.nortmann@web.de wrote:
"transient" (='t') is like "any", but requests that a variable should not be exported (ENV_FLAGS_VARACCESS_PREVENT_EXPORT).
"system" (='S') is meant for 'internal' variables that
The flags are positional, so 's' is not in use. It seems it would be cleaner to use a lower-case 's'.
You're probably right. I think back then I deliberately picked 'S' to avoid potential confusion (e.g. users specifying "s" when they actuall mean/want "ss"), as trying to set flags to just "S" would result in an error message.
For precisely that reason I'd actually prefer to find letter(s) that would not conflict with existing type or access flags, but 'v'olatile seemed ambiguous / too broad at that time.
Yes, that is a good point and was actually a consideration when I first chose the letters that are used for the current 2 positions. It's not a strict requirement, but it's less error-prone.
[...]
I'm not sure why you are adding "transient" or "volatile" to the varaccess. It is an orthogonal property of a variable. This is obvious from the fact that you need to add yet another to compose varaccess with varlifetime (or something).
I worked on something similar years ago, but never posted an RFC.
That's a very good point. This 'orthogonality' is what actually caused me to come up with that "transient" vs. "system" idea. I think I got misled by the way that U-Boot already combined various "access" bit flags, and it never occured to me to introduce another property with a third flag character. Actually "?av" (any-access, volatile) and "?rv" (read-only, volatile) would represent my intent perfectly well.
Perfect. That's exactly what I was proposing in 2010 (with the addition of auto-volatile).
I also like your "auto-volatile" concept a lot (i.e. resetting the volatile nature on setenv by the user). To stay unambiguous ('a' = "any" access), maybe this could be tagged 't'emporary.
't'emporary is ok, but I'm thinking something like o'v'erride(able) is more explicit about the behavior.
We'd also need a default letter for the lifesspan flag, possibly 'n'ormal?
I had suggested 'p'recious, but 'n'ormal is ok too. WD wanted a pointer type too, but I never got around to implementing it. Presumably that would conflict with 'p'recious if we went with that.
Again, it's not that important that the letters used be unique across positions, it's just less error prone. At the same time, using the first letter of a meaning more often will be less error-prone also.
The temporary (= auto-volatile) flag would also nicely save use from the need to have users fumble with "setenv .flags", and the quirks involved.
Exactly. It should be the uncommon case to need to touch the .flags.
Implementing this means some refactoring / a major overhaul of this RFC series, but I think it could be well worth that. I'll definitely give it a try when I find some time.
Awesome, thanks. -Joe