Re: Recommend we don't support users running built dojo on a git/hg/svn branch

On Sat, Jul 30, 2016 at 2:20 PM, David G <..hidden..> wrote:

Hi Chris,

In the past I would have to have agreed with you, however, in the last few days, ylavoie (irc/matrix) has implemented a git based technique to handle correct rebuilding of dojo only when needed for the travis tests.

I really think we need to avoid hardwiring git into the build process. For example, I don't use git (I use hg), which is why I pulled the git dependency out of make dojo. The make submodules is not a problem because mercurial handles that very well, but I already commented out the git checkout command in the make dojo section because I can't get dojo to build with it if I am not using git.

I believe the "exact" same technique can be used to resolve the issue on production installations as well.
However in a production environment, rather than "automatically" rebuilding, we can simply block login at the login page with a message that dojo modules need to be rebuilt.

Proliferating reasons why a user cannot log in may degrade the user experience quite a bit, I think. I really don't like the idea that we have a checklist of actions that must be handled after every pull -- it is not something I am comfortable handing off to an intermediately technical user as it will just cause frustration.

This actually fits in with some other changes I have planned for the login screen over the next 7 days, so I'll take ownership and prepare a PR that covers this as well.

Just don't break it for us Mercurial users.

Regards
David G

On 29/07/16 00:16, Chris Travers wrote:
Hi;

I spent a fair bit of time troubleshooting a bug that turned out to be a missing make dojo step. I know the user who reported it had trouble with the make dojo step and it didn't seem like a widget bug to me until I got fairly deep into it.

In the past we have supported people running LedgerSMB directly out of code repos or checkouts. I think we should continue to do that, but we really cannot guarantee that dojo gets rebuilt every time it should in these cases, so I think we should explicitly say that running with built dojo on a branch (stable or not) is not supported.

The built dojo strikes me as mostly useful when people are installing from packages and so we are providing the built libraries. But if we aren't providing the built libraries, then we can't be sure which versions they are using.

As a side note, i think we also need to really carefully think about how we can properly handle cache invalidation in our current approach.

Unless there are any objections, I will put a note saying that running with dojo_built=1 from git, hg, svn, etc is unsupported. I think that will avoid a lot of effort on our part and a lot of headache for would-be users.

--

Best Wishes,
Chris Travers

Efficito: Hosted Accounting and ERP. Robust and Flexible. No vendor lock-in.

http://www.efficito.com/learn_more
------------------------------------------------------------------------------
_______________________________________________
Ledger-smb-devel mailing list
..hidden..
https://lists.sourceforge.net/lists/listinfo/ledger-smb-devel
------------------------------------------------------------------------------

_______________________________________________
Ledger-smb-devel mailing list
..hidden..
https://lists.sourceforge.net/lists/listinfo/ledger-smb-devel