[Rpm-maint] [PATCH 1/2] Allow adding tag RPMTAG_AUTOINSTALLED to headers of package being installed

[Aleksei Nikiforov]

Hi

06.04.2018 15:39, Jeff Johnson пишет:
> 
> 
>> On Apr 6, 2018, at 8:14 AM, Aleksei Nikiforov <darktemplar at altlinux.org> wrote:
>>
>> Hi
>>
>> 06.04.2018 14:54, Jeff Johnson пишет:
>>> Sent from my iPad
>>>> On Apr 6, 2018, at 3:39 AM, Aleksei Nikiforov <darktemplar at altlinux.org> wrote:
>>>>
>>>> Hi
>>>>
>>>> 05.04.2018 23:17, Jeff Johnson пишет:
>>>>> Sent from my iPad
>>>>>> On Apr 4, 2018, at 9:25 AM, Aleksei Nikiforov <darktemplar at altlinux.org> wrote:
>>>>>>
>>>>>> Hi
>>>>>>
>>>>>> Since patch 2 is no longer needed, could you please provide a feedback about patch 1? It'd be great if this patch could be merged.
>>>>> I can try ...
>>>>> The patch seems (from examining the patch only with limited context) to be propagating AUTOINSTALL from the old -> new header at the rpmte layer in order to ensure that the appended AUTOINSTALL tag value is persistent across package upgrades.
>>>>> As before,  managing AUTOINSTALL persistence for a depsolver in rpmlib like this is rather awkward. E.g. what code should provide the add/modify/delete operations for AUTOINSTALL, particularly if/when the value is wrong?
>>>>> I suspect (from memory, I've not checked) that there are some odd corner cases, particularly if/when a package is reinstalled. But perhaps dropping AUTOINSTALL is a feature, not a bug, in that case.
>>>>> You likely should force a default value for AUTOINSTALL if/when the tag is not present because that is 1 fewer error returns to worry about. See how EPOCHNUM is implemented, likely as another header extension. You likely can use AUTOINSTALL as both an extension (mostly used by headerFormat) and as a tag-in-a-header (used by headerGet et al at the lower level) if you are careful.
>>>>
>>>> No, this patch is needed for adding this tag AUTOINSTALLED on package installation. When package is added to transaction via functions rpmtsAddInstallElement or rpmtsAddReinstallElement, it takes headers as one of arguments. But it doesn't save header itself, it saves just enough info to find package's file to read header from it later, when transaction would be running. Header passed to function is not saved, i.e. added/modified tags are not saved. Since rpm file doesn't contain AUTOINSTALLED tag, and shouldn't contain it due to the dynamic nature of it's value, the AUTOINSTALLED tag has to be saved there as well, so the value can be restored later.
>>> Ah, my bad: apologies for misanalyzing your patch (I'm on an iPad without grep, sigh).
>>>> So, this change only saves AUTOINSTALLED tag passed via functions rpmtsAddInstallElement and rpmtsAddReinstallElement. It's up to caller of those functions to add these headers, but if this tag isn't added function headerGetNumber returns 0 (i.e. sane default value). And later, when transaction is running, the value of this tag is restored from saved location and added to header read from file.
>>>>
>>>> Since on package upgrade or reinstall the header is read from file again, it doesn't contain tag AUTOINSTALLED, and adding this tag should be safe.
>>>>
>>>> Managing the value of this tag AUTOINSTALLED on package installation, reinstallation, upgrade or any other scenario is left to the user of library, except for setting to default value "zero" (meaning "manually installed") if no other value was provided.
>>>>
>>>>> Generally, I'd like to see RPM permit rpmlib users (like depsolvers) be able to freely append tags (arbitrary tagno, type, or reserved value) as needed without patching rpm itself.
>>>>
>>>> Well, yes, it's due to lack of a method to freely add tags to installed packages' headers or to mark additionally added headers to be copied from passed header to header saved into rpmdb (that's currently 2 different headers, even if they're similar) that I had to make this change.
>>> Well there is a set of methods that can be used to add AUTOINSTALL, just abstract and obscure.
>>> The RPMCALLBACK_* notify callback used to push progress bars can likely be extended to add AUTOINSTALL (or any other tag) before the header is saved in an rpmdb.
>>> Use rpmteHeader() to get the header within the callback, and pass the AUTOINSTALL (or any other tags to be added) state into the callback.
>>> See rpmShowProgress() in lib/rpminstall.c for what the rpm CLI does.
>>
>> Well, that looks like a hack I wouldn't want to implement.
>>
> 
> Hack? I am not suggesting patching lib/rpminstall.c to add AUTOINSTALL.
> 
> I am suggesting that you write a custom callback that adds AUTOINSTALL when a specific event is received, and pointed you at the CLI example. There is another example in rpm-python methods.
> 
> The advantage to the notify callback is that all known rpm depsolvers already are using the callback.
> 

Yes, I understood that it's proposed to not patch rpm library but use 
existing interfaces in such unexpected way instead. I just agree that 
proposed approach is obscure. I'd prefer to have more clean and 
straightforward API for this task.

>> If generic solution is desired for current change, it might be a good idea to modify functions rpmtsAddInstallElement and rpmtsAddReinstallElement and add new argument to these functions. Or add new versions of these functions and keep current ones for backwards compatibility. This argument would be some sort of container of tags and their values which would be added to header saved to rpmdb.
>>
> 
> There are many possible implementations that add a tag to a header somewhere, including your original patch.
>  > 73 de Jeff
>>> hth
>>> 73 de Jeff
>>>>> Yes, the header unload needs to be fixed to support more than just an int8 data type for full generality.
>>>>> hth
>>>>> 73 de Jeff
>>>>>> 28.03.2018 14:58, Aleksei Nikiforov пишет:
>>>>>>> Signed-off-by: Aleksei Nikiforov <darktemplar at altlinux.org>
>>>>>>> ---
>>>>>>>   lib/rpmte.c | 8 ++++++++
>>>>>>>   1 file changed, 8 insertions(+)
>>>>>>> diff --git a/lib/rpmte.c b/lib/rpmte.c
>>>>>>> index 40aa5e9..238c8b6 100644
>>>>>>> --- a/lib/rpmte.c
>>>>>>> +++ b/lib/rpmte.c
>>>>>>> @@ -39,6 +39,7 @@ struct rpmte_s {
>>>>>>>       char * arch;        /*!< Architecture hint. */
>>>>>>>       char * os;            /*!< Operating system hint. */
>>>>>>>       int isSource;        /*!< (TR_ADDED) source rpm? */
>>>>>>> +    uint32_t autoinstalled;    /*! Indicates whether package was installed just as dependency satisfier or not */
>>>>>>>         rpmte depends;              /*!< Package updated by this package (ERASE te) */
>>>>>>>       rpmte parent;        /*!< Parent transaction element. */
>>>>>>> @@ -191,6 +192,8 @@ static int addTE(rpmte p, Header h, fnpyKey key, rpmRelocation * relocs)
>>>>>>>       if (p->type == TR_ADDED)
>>>>>>>       p->pkgFileSize = headerGetNumber(h, RPMTAG_LONGSIGSIZE) + 96 + 256;
>>>>>>>   +    p->autoinstalled = headerGetNumber(h, RPMTAG_AUTOINSTALLED);
>>>>>>> +
>>>>>>>       rc = 0;
>>>>>>>     exit:
>>>>>>> @@ -576,6 +579,11 @@ static int rpmteOpen(rpmte te, int reload_fi)
>>>>>>>           rc = 1;
>>>>>>>       }
>>>>>>>       +    if (rc)
>>>>>>> +    {
>>>>>>> +        rc = (headerPutUint32(h, RPMTAG_AUTOINSTALLED, &(te->autoinstalled), 1) == 1);
>>>>>>> +    }
>>>>>>> +
>>>>>>>       rpmteSetHeader(te, h);
>>>>>>>       headerFree(h);
>>>>>>>       }
>>>>>>
>>>>>> Best regards
>>>>>> Aleksei Nikiforov
>>>>
>>>> Best regards
>>>> Aleksei Nikiforov
>>
>> Best regards
>> Aleksei Nikiforov

Best regards
Aleksei Nikiforov