Fix pg v7 regressions

[alitaheri]

This hacks around #7998.

I wrote a function which squashes the results returned by pg. except for when raw query doesn't have a type. In that case it returns whatever returned by pg (I don't think we should change this behavior).

This is a temporary fix until we come up with a solution that satisfies every use-case. Might require architectural changes. Since other dialects support multiple statements too.

Until then this fixes #7998

[alitaheri]

@felixfbecker The fields will come from the first result that contains them. The rows are the concatenation of all rows from all results. The rowCount is the accumulation of all non NaN rowCounts. And everything else command, parsers, etc. are from the last one. Anything out of place?

Also, when the query is raw with no type, the metadata might be an array ( possible breaking change ). But I'm not sure what others would expect of this. We explicitly notify the users that the results and metadata are dialect specific. so This is expected I guess. Right?

[oozimok]

Faced with this problem, tried to understand and find a solution. Later came across this merge request.

sequelize/lib/dialects/postgres/query.js

Array in queryResult:

My version of fixes:

.then(queryResults => {
  debug(`executed(${this.client.uuid || 'default'}) : ${this.sql}`);

  if (benchmark) {
    this.sequelize.log('Executed (' + (this.client.uuid || 'default') + '): ' + this.sql, Date.now() - queryBegin, this.options);
  }

  return queryResults;
})
.then(queryResults => {
  const queryResult = Array.isArray(queryResults) ? queryResults.pop() : queryResults;
  const rows = queryResult.rows;
  const rowCount = queryResult.rowCount;

node v7.7.3, npm v5.3.0, sequelize v4.4.2, pg v7.0.2

P.S. At the moment rolled back to the 6th version of pg. The flight is normal.

[alitaheri]

@4e1ovek That won't always work. Some queries might have their main stuff in the second statement.

[codecov]

Codecov Report

Merging #8029 into master will increase coverage by <.01%.
The diff coverage is 100%.

[corysimmons]

I don't mean to be one of those entitled devs, but can contribs/maintainers take this one a bit more seriously and get a fix, or at least a documented workaround, merged?

Broken enums is a huge problem.

[alitaheri]

@corysimmons downgrade to pg 6 for now. it's not just the enums that are broken. Any multi-statement query will fail.

[corysimmons]

@alitaheri So pg @ 6.4.2. Do I need to add pg-native? Do I need to downgrade pg-hstore to something?

[alitaheri]

@corysimmons You don't need pg-native. Best not use it anyway, some features aren't implemented there. And pg-hstore is a simple serializer/deserializer with no dependencies. Just install pg@6.4.2 and your issue will be fixed for now.

P.S. pg and pg-native are installed here only for testing purposes.

[corysimmons]

Thanks Ali. Maybe we can update the README with a warning about pg7 while a fix is in the works?

[gabegorelick]

Is there anything that needs to be done to get this into a mergable state (besides rebasing package.json)?

[joaovitoras]

Do you intend to merge this pull request?