inspectdb and field names ending with underscores

[3gun <mihail.lukin@…>]

Hi,

There is a field named 'type_' in mysql table. Inspectdb inspects it as 'type_' and model does not validate. I suggest removing underscores from model field name and specifying column_db argument in field constructor.

< type_ = models.CharField(max_length=90, db_column='type_')

type = models.CharField(max_length=90, db_column='type_')

Sorry for posting without login. I've forgot my password and didn't find restoring feature on this site.

[Russell Keith-Magee]

I've confirmed that the problem exists, but your proposed solution won't work (at least, not in the general case). type_ is the name of a Python builtin, so you shouldn't name a model field type. An alternate approach would be to append or prepend a known character (e.g., xtype, type_1) to create a field name.

[Alex Gaynor]

Django already handles the case of a python keyword: ​http://code.djangoproject.com/browser/django/trunk/django/core/management/commands/inspectdb.py#L64 . Having something with the same name as a builtin shouldn't matter.

[Elijah Rutschman]

There are a few other scenarios where invalid field names could be generated via the inspectdb command. For instance, a column name of "Foo/Bar", while unconventional, is a valid column name, at least in MySQL 5, but the invalid field name of 'foo/bar' would be generated.

Currently, inspectdb ensures that:

• field names do not contain spaces or dashes
• field names are not Python keywords

The patch I just attached improves this by also ensuring that:

• field names do not contain any characters outside of a-zA-Z0-9_
• field names do not contain double underscores
• field names neither start nor end with underscores
• field names do not start with digits

[Elijah Rutschman]

Patch that addresses trailing underscore issue as well as several other introspection field naming scenarios

[kgibula]

Patch can't be applied to trunk anymore. It would also convert both ''Foo__'' and ''foo_'' (valid MySQL column names) to ''foo_field''.

[kgibula]

updated patch

[kgibula]

I tested it on MySQL, PostgreSQL and SQLite. Currently tests pass only on MySQL and PostgreSQL. When db_column keyword argument is added, on MySQL and PostgreQSL it looks like this: db_column='name_of_field'. On SQLite however it looks like this: db_column=u'name_of_field'. I am not sure which one is correct.

[Claude Paroz]

Updated patch

[Claude Paroz]

Here's an updated patch where I have slightly refactored things to isolate the normalization process of the field name. I purposely did not include the non-ascii characters issue which was also addressed by previous "inspectdb.diff" patch. If the committer finds it's better to also include it, I can complete the patch. Otherwise, I'll gladly add a specific patch for that issue in #16737 after this one is committed.

[Claude Paroz]

Patch updated to current trunk

[Claude Paroz <claude@…>]

In [395c6083af93cc37c7d16ef4db451091841cefdc]:

Fixed #12460 -- Improved inspectdb handling of special field names

Thanks mihail lukin for the report and elijahr and kgibula for their
work on the patch.