[PATCH] Making exported variables really global

[PATCH] Making exported variables really global

[Andreas Vogel]

[-- Attachment #1: Type: text/plain, Size: 578 bytes --]

Hi all,

This tiny patch makes exported variables really global. Right now
exported variables are set in the context of a submenu when running it.
But any changes to those vars are lost when leaving the submenu. This
patch sets and exports all vars in the calling context for the submenu
which are exported in the context of the submenu.

This feature is needed in order to have submenus implementing something
like setting boot options like boot params, language setttings and the
like using environment variables.

Any comments? If it's OK any chance to be included?

Andreas

[-- Attachment #2: 08-global_environ_vars.patch --]
[-- Type: text/plain, Size: 800 bytes --]

=== modified file 'grub-core/normal/context.c'
--- grub-core/normal/context.c	2012-03-01 21:43:03 +0000
+++ grub-core/normal/context.c	2012-03-04 13:51:28 +0000
@@ -126,6 +126,20 @@
 
       for (p = grub_current_context->vars[i]; p; p = q)
 	{
+	  if (p->global)
+	    {
+	      /* Set and export all global variables inside
+		 the calling/previous context.  */
+	      struct grub_env_context *tmp_context = grub_current_context;
+	      grub_current_context = grub_current_context->prev;
+	      if (grub_env_set (p->name, p->value) == GRUB_ERR_NONE)
+		{
+		  grub_env_export (p->name);
+		  grub_register_variable_hook (p->name, p->read_hook, p->write_hook);
+		}
+	      grub_current_context = tmp_context;
+	    }
+
 	  q = p->next;
           grub_free (p->name);
 	  grub_free (p->value);

Re: [PATCH] Making exported variables really global

[Vladimir 'φ-coder/phcoder' Serbinenko]

On 04.03.2012 14:08, Andreas Vogel wrote:
> Hi all,
>
> This tiny patch makes exported variables really global. Right now
> exported variables are set in the context of a submenu when running it.
> But any changes to those vars are lost when leaving the submenu. This
> patch sets and exports all vars in the calling context for the submenu
> which are exported in the context of the submenu.
We follow the bash behaviour. And the bash behaviour is exactly what 
GRUB does:
phcoder@debian.x201.phnet:14:15:11:~/grub2/bzr/mainline$ bash
phcoder@debian.x201.phnet:14:15:15:~/grub2/bzr/mainline$ export XYZ=x
phcoder@debian.x201.phnet:14:15:19:~/grub2/bzr/mainline$ exit
phcoder@debian.x201.phnet:14:15:20:~/grub2/bzr/mainline$ echo $XYZ

phcoder@debian.x201.phnet:14:15:24:~/grub2/bzr/mainline$

-- 
Regards
Vladimir 'φ-coder/phcoder' Serbinenko

Re: [PATCH] Making exported variables really global

[Andreas Vogel]

Am 04.03.2012 14:18, schrieb Vladimir 'φ-coder/phcoder' Serbinenko:
> On 04.03.2012 14:08, Andreas Vogel wrote:
>> Hi all,
>>
>> This tiny patch makes exported variables really global. Right now
>> exported variables are set in the context of a submenu when running it.
>> But any changes to those vars are lost when leaving the submenu. This
>> patch sets and exports all vars in the calling context for the submenu
>> which are exported in the context of the submenu.
> We follow the bash behaviour. And the bash behaviour is exactly what
> GRUB does:
> phcoder@debian.x201.phnet:14:15:11:~/grub2/bzr/mainline$ bash
> phcoder@debian.x201.phnet:14:15:15:~/grub2/bzr/mainline$ export XYZ=x
> phcoder@debian.x201.phnet:14:15:19:~/grub2/bzr/mainline$ exit
> phcoder@debian.x201.phnet:14:15:20:~/grub2/bzr/mainline$ echo $XYZ
>
> phcoder@debian.x201.phnet:14:15:24:~/grub2/bzr/mainline$
>
I understand your point, but IMHO we do need to be able to set global
vars in submenus, that's my patch about.

Otherwise how can you use a "Settings..." submenu to configure
environment variables which are used by other menu entries for booting?
The analogy with bash is not useful here.

Re: [PATCH] Making exported variables really global

[Vladimir 'φ-coder/phcoder' Serbinenko]

On 04.03.2012 14:47, Andreas Vogel wrote:
> Am 04.03.2012 14:18, schrieb Vladimir 'φ-coder/phcoder' Serbinenko:
>> On 04.03.2012 14:08, Andreas Vogel wrote:
>>> Hi all,
>>>
>>> This tiny patch makes exported variables really global. Right now
>>> exported variables are set in the context of a submenu when running it.
>>> But any changes to those vars are lost when leaving the submenu. This
>>> patch sets and exports all vars in the calling context for the submenu
>>> which are exported in the context of the submenu.
>> We follow the bash behaviour. And the bash behaviour is exactly what
>> GRUB does:
>> phcoder@debian.x201.phnet:14:15:11:~/grub2/bzr/mainline$ bash
>> phcoder@debian.x201.phnet:14:15:15:~/grub2/bzr/mainline$ export XYZ=x
>> phcoder@debian.x201.phnet:14:15:19:~/grub2/bzr/mainline$ exit
>> phcoder@debian.x201.phnet:14:15:20:~/grub2/bzr/mainline$ echo $XYZ
>>
>> phcoder@debian.x201.phnet:14:15:24:~/grub2/bzr/mainline$
>>
> I understand your point, but IMHO we do need to be able to set global
> vars in submenus, that's my patch about.
>
> Otherwise how can you use a "Settings..." submenu to configure
> environment variables which are used by other menu entries for booting?
> The analogy with bash is not useful here.
We can add a command with traits of both "menuentry" (same context, same 
menu) and "submenu" (new context, new menu), sth like submemu_source 
(same context, new menu)
>


-- 
Regards
Vladimir 'φ-coder/phcoder' Serbinenko

Re: [PATCH] Making exported variables really global

[Andreas Vogel]

Am 04.03.2012 15:10, schrieb Vladimir 'φ-coder/phcoder' Serbinenko:
> On 04.03.2012 14:47, Andreas Vogel wrote:
>> Am 04.03.2012 14:18, schrieb Vladimir 'φ-coder/phcoder' Serbinenko:
>>> On 04.03.2012 14:08, Andreas Vogel wrote:
>>>> Hi all,
>>>>
>>>> This tiny patch makes exported variables really global. Right now
>>>> exported variables are set in the context of a submenu when running
>>>> it.
>>>> But any changes to those vars are lost when leaving the submenu. This
>>>> patch sets and exports all vars in the calling context for the submenu
>>>> which are exported in the context of the submenu.
>>> We follow the bash behaviour. And the bash behaviour is exactly what
>>> GRUB does:
>>> phcoder@debian.x201.phnet:14:15:11:~/grub2/bzr/mainline$ bash
>>> phcoder@debian.x201.phnet:14:15:15:~/grub2/bzr/mainline$ export XYZ=x
>>> phcoder@debian.x201.phnet:14:15:19:~/grub2/bzr/mainline$ exit
>>> phcoder@debian.x201.phnet:14:15:20:~/grub2/bzr/mainline$ echo $XYZ
>>>
>>> phcoder@debian.x201.phnet:14:15:24:~/grub2/bzr/mainline$
>>>
>> I understand your point, but IMHO we do need to be able to set global
>> vars in submenus, that's my patch about.
>>
>> Otherwise how can you use a "Settings..." submenu to configure
>> environment variables which are used by other menu entries for booting?
>> The analogy with bash is not useful here.
> We can add a command with traits of both "menuentry" (same context,
> same menu) and "submenu" (new context, new menu), sth like
> submemu_source (same context, new menu) 
Yes, that's one possible solution.

Why establishing a new command instead of using an option for submenu,
let's say "--export 0|1"?

Re: [PATCH] Making exported variables really global

[Andreas Vogel]

Am 04.03.2012 17:16, schrieb Andreas Vogel:
> Am 04.03.2012 15:10, schrieb Vladimir 'φ-coder/phcoder' Serbinenko:
>> On 04.03.2012 14:47, Andreas Vogel wrote:
>>> Am 04.03.2012 14:18, schrieb Vladimir 'φ-coder/phcoder' Serbinenko:
>>>> On 04.03.2012 14:08, Andreas Vogel wrote:
>>>>> Hi all,
>>>>>
>>>>> This tiny patch makes exported variables really global. Right now
>>>>> exported variables are set in the context of a submenu when running
>>>>> it.
>>>>> But any changes to those vars are lost when leaving the submenu. This
>>>>> patch sets and exports all vars in the calling context for the submenu
>>>>> which are exported in the context of the submenu.
>>>> We follow the bash behaviour. And the bash behaviour is exactly what
>>>> GRUB does:
>>>> phcoder@debian.x201.phnet:14:15:11:~/grub2/bzr/mainline$ bash
>>>> phcoder@debian.x201.phnet:14:15:15:~/grub2/bzr/mainline$ export XYZ=x
>>>> phcoder@debian.x201.phnet:14:15:19:~/grub2/bzr/mainline$ exit
>>>> phcoder@debian.x201.phnet:14:15:20:~/grub2/bzr/mainline$ echo $XYZ
>>>>
>>>> phcoder@debian.x201.phnet:14:15:24:~/grub2/bzr/mainline$
>>>>
>>> I understand your point, but IMHO we do need to be able to set global
>>> vars in submenus, that's my patch about.
>>>
>>> Otherwise how can you use a "Settings..." submenu to configure
>>> environment variables which are used by other menu entries for booting?
>>> The analogy with bash is not useful here.
>> We can add a command with traits of both "menuentry" (same context,
>> same menu) and "submenu" (new context, new menu), sth like
>> submemu_source (same context, new menu) 
> Yes, that's one possible solution.
>
> Why establishing a new command instead of using an option for submenu,
> let's say "--export 0|1"?

Please let me summarize this issue after thinking about it again:

Right now the submenu command opens a new context. All exported vars in
the parent context are also exported to the new context of the submenu.
This is the same behaviour like for Unix shells. The menuentry command
doesn't open a new context, it runs in the context of the caller.

For submenus there are 3 possibilities for how to behave:

1) Create a new context and run the submenu in the new context. After
exiting the submenu, all variables in the submenu context are lost (this
is the current behavior).

2) Don't create a new context and run the submenu in the same context as
the caller. This would be the same behavior as for menuentry commands.
This can be implemented by having a separate submenu command (e.g.
submenu_source, or submenu_shared) or introducing a new submenu option
(e.g. --shared 0| 1).

3) Create a new context for the submenu. On exit all exported
environment variables in the submenu context are exported to the context
of the caller. By this you have the advantage that all local
(unexported) variables in the submenu context are really local, even for
further submenu entries in the submenu, but all exported variables are
global for the whole menu tree. Like for option 2) this can be achieved
by establishing a new command or using an new option for the submenu
command. But this behavior can also be the default behavior for submenus
without the need to be able to enable/disable it.


As I'm quite unfamiliar with the internal design of GRUB, I'm not sure
what option is most adequate, even though I'd vote for option 3).

I cannot see the point why it's so important to run a submenu in its own
shell (= context) and why menuentries of a submenu shouldn't be able to
access the same environment variables like the menuentries of a parent
menu. In my opinion the analogy to shells and sub-shells in Unix isn't
really adequate because for me a menu tree is an entity on its own,
regardless of being implemented as commands (like in GRUB) or using some
kind of configuration file defining the menu tree.
On the other side i see some advantages for submenus having their own
context, e.g. to have local environment variables which don't 'pollute'
the calling context. That's the reason why I vote for option 3) with the
addition that it's not

Please excuse my lengthy exploration about this issue, but it's a very
important point for GRUB. Not being able to set globally defined
environment variables (= exported toplevel context vars) in submenus
prevented until now having customization submenus and doing a search in
I-Net about this issue, I've found a lot of posts desperately asking for
this feature.

As you can see from my patch in the OP, the modification is quite simple
and from what i can see the implementation of the other options
presented above are not much more complicated.

I'm willing to implement and supply whatever will be decided in the hope
that we can get it into the upcoming version. We just need to get a
final decision. So any comments are, as always, highly welcomed.

Andreas

Re: [PATCH] Making exported variables really global

[Andreas Born]

Am 04.03.2012 18:55, schrieb Andreas Vogel:
> Please let me summarize this issue after thinking about it again:
>
> Right now the submenu command opens a new context. All exported vars in
> the parent context are also exported to the new context of the submenu.
> This is the same behaviour like for Unix shells. The menuentry command
> doesn't open a new context, it runs in the context of the caller.
>
> For submenus there are 3 possibilities for how to behave:
>
> 1) Create a new context and run the submenu in the new context. After
> exiting the submenu, all variables in the submenu context are lost (this
> is the current behavior).
Personally, I like that solution best.
What's the problem? Currently there's no way and absolutely no way to 
reliably get data from a submenu to a parent menu. This makes it 
impossible to use submenus e.g. for settings menus.
By sharing the context this would not only be consistent with 
menuentries but while significantly improving the situation it can't 
cause any major harm. In case somebody relies on variables being local, 
I'm pretty sure there's always a solution to achieve similar behaviour 
by multiple variables. So making the context shared wouldn't create any 
real troubles. But the way it is now - separate context for submenu - 
does create problems. For example no settings menus and there's no 
solution for that need at all! As in my opinion this is a major 
applications for submenus something should be done about it.
> 2) Don't create a new context and run the submenu in the same context as
> the caller. This would be the same behavior as for menuentry commands.
> This can be implemented by having a separate submenu command (e.g.
> submenu_source, or submenu_shared) or introducing a new submenu option
> (e.g. --shared 0| 1).
While I could live with such a solution, it seems just flawed to me. 
Such a construct introduces unecessary complexity and inconsistency. If 
there's such a switch for submenus why shouldn't there be one for 
source, menuentry, ...?
As pointed out previously a separated context offers pretty much nothing 
that's not possible with a shared one, so in most cases --shared 1 would 
be fine. But having to write every time --shared 1 after a submenu or 
submenu_shared is quite cluttered. The use of such constructs doesn't 
create the impression of a well designed language for my part either.
> 3) Create a new context for the submenu. On exit all exported
> environment variables in the submenu context are exported to the context
> of the caller. By this you have the advantage that all local
> (unexported) variables in the submenu context are really local, even for
> further submenu entries in the submenu, but all exported variables are
> global for the whole menu tree. Like for option 2) this can be achieved
> by establishing a new command or using an new option for the submenu
> command. But this behavior can also be the default behavior for submenus
> without the need to be able to enable/disable it.
I don't really like this one. It introduces a completely new and 
complicated behaviour, which is not commonly found among other software, 
especially not with bash.
I see the risk that even with very good documentation, this 
implementation leads to unexpected results. And I think it could happen 
that users won't be able to understand what's causing the problems 
because shared and local variables are mixed. Such unintuitive solutions 
should be avoided wherever possible and there are two other 
possibilities which are in my view much better.


Anyway, just some thoughts from a user.

Andreas

Re: [PATCH] Making exported variables really global

[Andreas Vogel]

First of all thanks for your detailed comment.

Am 04.03.2012 22:03, schrieb Andreas Born:
> Am 04.03.2012 18:55, schrieb Andreas Vogel:
>> Please let me summarize this issue after thinking about it again:
>>
>> Right now the submenu command opens a new context. All exported vars in
>> the parent context are also exported to the new context of the submenu.
>> This is the same behaviour like for Unix shells. The menuentry command
>> doesn't open a new context, it runs in the context of the caller.
>>
>> For submenus there are 3 possibilities for how to behave:
>>
>> 1) Create a new context and run the submenu in the new context. After
>> exiting the submenu, all variables in the submenu context are lost (this
>> is the current behavior).
> Personally, I like that solution best.
> What's the problem? Currently there's no way and absolutely no way to
> reliably get data from a submenu to a parent menu. This makes it
> impossible to use submenus e.g. for settings menus.
> By sharing the context this would not only be consistent with
> menuentries but while significantly improving the situation it can't
> cause any major harm. In case somebody relies on variables being
> local, I'm pretty sure there's always a solution to achieve similar
> behaviour by multiple variables. So making the context shared wouldn't
> create any real troubles. But the way it is now - separate context for
> submenu - does create problems. For example no settings menus and
> there's no solution for that need at all! As in my opinion this is a
> major applications for submenus something should be done about it.
It seems you are missing the point here. Option 1) describes the
situation like it is right now and means not to change the code and not
being able to have submenus which are able to set variables in the
callers context. As far as I understood Vladimirs first answer, there is
already some kind of common understanding that this is something what
should be changed in the future.

>> 2) Don't create a new context and run the submenu in the same context as
>> the caller. This would be the same behavior as for menuentry commands.
>> This can be implemented by having a separate submenu command (e.g.
>> submenu_source, or submenu_shared) or introducing a new submenu option
>> (e.g. --shared 0| 1).
> While I could live with such a solution, it seems just flawed to me.
> Such a construct introduces unecessary complexity and inconsistency.
> If there's such a switch for submenus why shouldn't there be one for
> source, menuentry, ...?
> As pointed out previously a separated context offers pretty much
> nothing that's not possible with a shared one, so in most cases
> --shared 1 would be fine. But having to write every time --shared 1
> after a submenu or submenu_shared is quite cluttered. The use of such
> constructs doesn't create the impression of a well designed language
> for my part either.
As far as I understood your comment, you're voting for option 2) but
without needing to specify that the context should be shared with the
calling context. In other words sharing the context should be enabled by
default. Right?

>> 3) Create a new context for the submenu. On exit all exported
>> environment variables in the submenu context are exported to the context
>> of the caller. By this you have the advantage that all local
>> (unexported) variables in the submenu context are really local, even for
>> further submenu entries in the submenu, but all exported variables are
>> global for the whole menu tree. Like for option 2) this can be achieved
>> by establishing a new command or using an new option for the submenu
>> command. But this behavior can also be the default behavior for submenus
>> without the need to be able to enable/disable it.
> I don't really like this one. It introduces a completely new and
> complicated behaviour, which is not commonly found among other
> software, especially not with bash.
> I see the risk that even with very good documentation, this
> implementation leads to unexpected results. And I think it could
> happen that users won't be able to understand what's causing the
> problems because shared and local variables are mixed. Such
> unintuitive solutions should be avoided wherever possible and there
> are two other possibilities which are in my view much better.
The rule for this option is very simple and can be expressed in one
sentence: "Regarding the menu system, all exported variables have a
global scope."
BTW, right now GRUB makes a difference between local (unexported) and
global (exported) vars: global vars are "handed over" to the context of
submenus, local vars not. Options 3) just enhances this paradigm and
treats global vars as read/write instead of readonly.
The comparison with bash is senseless - we are talking about a menu
system and not about subshells in general.

Andreas

Re: [PATCH] Making exported variables really global

[Andreas Born]

Am 04.03.2012 23:48, schrieb Andreas Vogel:
> First of all thanks for your detailed comment.
>
> Am 04.03.2012 22:03, schrieb Andreas Born:
>> Am 04.03.2012 18:55, schrieb Andreas Vogel:
>>> Please let me summarize this issue after thinking about it again:
>>>
>>> Right now the submenu command opens a new context. All exported vars in
>>> the parent context are also exported to the new context of the submenu.
>>> This is the same behaviour like for Unix shells. The menuentry command
>>> doesn't open a new context, it runs in the context of the caller.
>>>
>>> For submenus there are 3 possibilities for how to behave:
>>>
>>> 1) Create a new context and run the submenu in the new context. After
>>> exiting the submenu, all variables in the submenu context are lost (this
>>> is the current behavior).
>> Personally, I like that solution best.
>> What's the problem? Currently there's no way and absolutely no way to
>> reliably get data from a submenu to a parent menu. This makes it
>> impossible to use submenus e.g. for settings menus.
>> By sharing the context this would not only be consistent with
>> menuentries but while significantly improving the situation it can't
>> cause any major harm. In case somebody relies on variables being
>> local, I'm pretty sure there's always a solution to achieve similar
>> behaviour by multiple variables. So making the context shared wouldn't
>> create any real troubles. But the way it is now - separate context for
>> submenu - does create problems. For example no settings menus and
>> there's no solution for that need at all! As in my opinion this is a
>> major applications for submenus something should be done about it.
> It seems you are missing the point here. Option 1) describes the
> situation like it is right now and means not to change the code and not
> being able to have submenus which are able to set variables in the
> callers context. As far as I understood Vladimirs first answer, there is
> already some kind of common understanding that this is something what
> should be changed in the future.
>
>>> 2) Don't create a new context and run the submenu in the same context as
>>> the caller. This would be the same behavior as for menuentry commands.
>>> This can be implemented by having a separate submenu command (e.g.
>>> submenu_source, or submenu_shared) or introducing a new submenu option
>>> (e.g. --shared 0| 1).
>> While I could live with such a solution, it seems just flawed to me.
>> Such a construct introduces unecessary complexity and inconsistency.
>> If there's such a switch for submenus why shouldn't there be one for
>> source, menuentry, ...?
>> As pointed out previously a separated context offers pretty much
>> nothing that's not possible with a shared one, so in most cases
>> --shared 1 would be fine. But having to write every time --shared 1
>> after a submenu or submenu_shared is quite cluttered. The use of such
>> constructs doesn't create the impression of a well designed language
>> for my part either.
> As far as I understood your comment, you're voting for option 2) but
> without needing to specify that the context should be shared with the
> calling context. In other words sharing the context should be enabled by
> default. Right?
Yes. Somehow missed that 1) wasn't about this. Sorry. Let's call it 4), 
which would be: no new context is created with submenus.
>
>>> 3) Create a new context for the submenu. On exit all exported
>>> environment variables in the submenu context are exported to the context
>>> of the caller. By this you have the advantage that all local
>>> (unexported) variables in the submenu context are really local, even for
>>> further submenu entries in the submenu, but all exported variables are
>>> global for the whole menu tree. Like for option 2) this can be achieved
>>> by establishing a new command or using an new option for the submenu
>>> command. But this behavior can also be the default behavior for submenus
>>> without the need to be able to enable/disable it.
>> I don't really like this one. It introduces a completely new and
>> complicated behaviour, which is not commonly found among other
>> software, especially not with bash.
>> I see the risk that even with very good documentation, this
>> implementation leads to unexpected results. And I think it could
>> happen that users won't be able to understand what's causing the
>> problems because shared and local variables are mixed. Such
>> unintuitive solutions should be avoided wherever possible and there
>> are two other possibilities which are in my view much better.
> The rule for this option is very simple and can be expressed in one
> sentence: "Regarding the menu system, all exported variables have a
> global scope."
Yes, but no matter where and how prominent you put it who reads this?
And not introducing the complexity of contexts with subshells, which 
offer little benefit here, makes it even easier.
> BTW, right now GRUB makes a difference between local (unexported) and
> global (exported) vars: global vars are "handed over" to the context of
> submenus, local vars not.
That's the same distinction subshells make so it's a common and 
well-known concept.
>   Options 3) just enhances this paradigm and
> treats global vars as read/write instead of readonly.
We actually then have three cases (there are 2 with 4) and even with 2) ):
  1) local variables: only visible in the current context
  2) exported/global variables: copied to new contexts
  3) really global variables: copied back and forth between contexts
This is unnecessarily complicated and using 3) only with a special 
command like submenu doesn't make the distinction easier to grasp.
In my view this is even easier:
"In the menu system there is one shared context."
And this behaves like menuentries.
> The comparison with bash is senseless - we are talking about a menu
> system and not about subshells in general.
Bash and it's concepts are well known. On the other hand introducing 
new, not commonly used concepts makes it unnecessarily hard to use and 
understand the grub shell. Why make it complicated if you can keep it 
simple?

Andreas