Allow sys.stdout to be replaced

[alanjds]

Please refer to the discussion on #139

Do not merge until __getattr__ and __setattr__ is working.

[alanjds]

Comment by corona10
Friday Apr 28, 2017 at 15:51 GMT

---

I have one thing to discuss here.
At this point, I think that we should create module directory for these kinds of task. runtime/sysmodule.go => module/sysmodule.go
WDYT? @trotterdylan @alanjds

[alanjds]

Comment by alanjds
Friday Apr 28, 2017 at 16:25 GMT

---

I totally agree, but am not mature enough on Golang to glance the implications of this change.

[alanjds]

Comment by corona10
Friday Apr 28, 2017 at 16:27 GMT

---

I believe that more modules will be needed. e.g (socketModule etc..)

[alanjds]

Comment by alanjds
Saturday Apr 29, 2017 at 20:55 GMT

---

This is no more a WIP, after changed from __getattr__ to __getattribute__ fixed it.

Will continue the #139 implementation over this code.

Please discuss what changes we want before this (#146) got be merged.

[alanjds]

Comment by trotterdylan
Saturday Apr 29, 2017 at 23:28 GMT

---

I would avoid __getattribute__ and instead use __getattr__. The latter is called only after nothing is found through normal attribute lookup, which is probably what we want in this case.

[alanjds]

Comment by alanjds
Sunday Apr 30, 2017 at 01:48 GMT

---

Me too. But this rabbit hole is getting too deep!
Let me close some Print() issues first and later I come back to implement __getattr__.

This __getattr__ issue seems at least one magnitude harder than the other issues anyway. I will need to get "slots" understanding first.

[alanjds]

Comment by alanjds
Monday May 01, 2017 at 19:00 GMT

---

Just to say, tracebacks would be very nice. Took me quite a bit to realize this 'res' typo

[alanjds]

Comment by alanjds
Saturday Jun 10, 2017 at 15:48 GMT

---

I would avoid getattribute and instead use getattr. The latter is called only after nothing is found through normal attribute lookup, which is probably what we want in this case.

@trotterdylan Are you ok with going on this way, and allow to use getattribute for now, or is better to go develop getattr before keep going on this PR?

[alanjds]

Comment by trotterdylan
Wednesday Jun 14, 2017 at 14:35 GMT

---

I think that __getattr__ should be fairly straightforward to implement. It needs a new slot and fallback logic in grumpy.GetAttr(). Also a few tests that validate the correct attribute resolution.

[alanjds]

Comment by alanjds
Friday Jun 16, 2017 at 16:40 GMT

---

Ok. Then I will chase __getattr__ on another PR.

Everything else is ok?

[alanjds]

Comment by trotterdylan
Monday Jun 19, 2017 at 15:05 GMT

---

I think this is a little more complicated than I had hoped it would be. What if we get rid of the hybrid module concept and instead swap out sys.__dict__ something like this:

# lib/sys.py
from __go__.grumpy import SysDict, SysModules
SysModules['sys'].__dict__ = SysDict
... carry on with the module as normal

Currently __dict__ is not settable, but it will be once google#331 is merged.

[alanjds]

Comment by alanjds
Monday Jun 19, 2017 at 18:46 GMT

---

Hum... the whole point of this PR is allow sys.stdout = StringIO() to propagate to print.

If SysModules['sys'].__dict__ = SysDict fixes this need and is simpler, lets do it!

Seems to still need an [SysDict[k] = locals()[k] for k in __all__] at the last line, but for me it looks more ok than the HybridModule stuff.

[alanjds]

Comment by trotterdylan
Monday Jun 19, 2017 at 19:58 GMT

---

Right, so the trick would be get rid of the Stdout, Stdin and Stderr grumpy package members and instead pull those from SysDict like SysDict.GetItemString(f, "stdin"). I have a change started to see how this would work in practice. I can either finish that up and create a PR, or if you like, you can go that route in this PR.

Seems to still need an [SysDict[k] = locals()[k] for k in all] at the last line, but for me it looks more ok than the HybridModule stuff.

I don't think this is necessary if we do the __dict__ assignment at the very top of the module.