bigquery: support SQL parameters

[jba]

See parameterMode and queryParameters in the query job reference.

[c0b]

+1 on this

since https://googlecloudplatform.github.io/google-cloud-node/#/docs/bigquery/0.4.0/bigquery?method=query allows any additional parameters in the options, we can put parameterMode and queryParameters in the options,

but the problem is even the reference isn't very clear,
https://cloud.google.com/bigquery/docs/reference/rest/v2/jobs/query#request-body

this is what it says, but what are the possible values for the string typed parameterMode? like positional or named ?

parameterMode string [Experimental] Standard SQL only. Whether to use positional (?) or named (@myparam) query parameters in this query.

if I choose named what's the syntax I can use in the query?

query string [Required] A query string, following the BigQuery query syntax, of the query to execute. Example: "SELECT count(f1) FROM [myProjectId:myDatasetId.myTableId]".

how to request additional elaboration on the reference itself?

[c0b]

I've figured out an example how to run without special support from this library, from the python bg tool's apilog (from the google-cloud-sdk), also from ./opt/google-cloud-sdk/platform/bq/bq.py source code (googleapis/google-cloud-python#2342 (comment) says the google-cloud-sdk is not open source, but you can read the source to see how queryParameters is constructed):

$ bq --apilog - query --use_legacy_sql=false --parameter arr:'ARRAY<INTEGER>':'[2,3,4]' 'SELECT 3 IN UNNEST(@arr)'
[...]

INFO:root:body: {"configuration": {"query": {"query": "SELECT 3 IN UNNEST(@arr)", "useLegacySql": false, "queryParameters": [{"parameterType": {"type": "ARRAY", "arrayType": {"type": "INTEGER"}}, "parameterValue": {"arrayValues": [{"value": 2}, {"value": 3}, {"value": 4}]}, "name": "arr"}]}}, "jobReference": ...}

INFO:root:{
 "kind": "bigquery#job",
 "etag": "...",
 "id": "...",
 "selfLink": "...",
 "jobReference": {
    ...
 },
 "configuration": {
  "query": {
   "query": "SELECT 3 IN UNNEST(@arr)",
   "destinationTable": {
      ...
   },
   "createDisposition": "CREATE_IF_NEEDED",
   "writeDisposition": "WRITE_TRUNCATE",
   "useLegacySql": false,
   "queryParameters": [
    {
     "name": "arr",
     "parameterType": {
      "type": "ARRAY",
      "arrayType": {
       "type": "INT64"
      }
     },
     "parameterValue": {
      "arrayValues": [
       {
        "value": "2"
       },
       {
        "value": "3"
       },
       {
        "value": "4"
       }
      ]
     }
    }
   ]
  }
 },
 "status": {
  "state": "RUNNING"
 },
 "statistics": {
  "creationTime": "1479104934691",
  "startTime": "1479104934979"
 },
}
INFO:root:--response-end--
+------+
| f0_  |
+------+
| true |
+------+

this is saying:

1. parameterMode is optional: I can set it to 'positional' or 'named' but any other value results an error, but if not sending this, it's totally fine; you see bq tool is not sending parameterMode at all
2. it's also a good example of how ARRAY type should be sent, it's in a recursive structure, from this I inferred how to construct a STRUCT type and it works

or saying if I am doing queryParameters in this structure on myself, that just works fine, but maintainers of this library can make an API for users to use queryParameters easier

   "queryParameters": [
    {
     "name": "arr",
     "parameterType": {
      "type": "ARRAY",
      "arrayType": {
       "type": "INT64"
      }
     },
     "parameterValue": {
      "arrayValues": [
       {
        "value": "2"
       },
       {
        "value": "3"
       },
       {
        "value": "4"
       }
      ]
     }

[jba]

@c0b, thanks for your explorations! They have been a big help as we start to implement this.

[jba]

This is our design for query parameters, which is implemented in CLs 9557, 9558, 9559 and 9561:

As shown in c0b's comments above, we can ignore parameter mode. Also, specifying a query parameter requires a value, expressed in a particular way that doesn't match anything else in the API, and a corresponding type, which also has a unique way of being expressed.

Instead of requiring users to provide values and types in this way, we accept query parameters as ordinary Go values. We use reflection to encode the Go value as a query parameter value, as well as to determine and encode its type.

We can easily deal with most BigQuery types this way:

• The various flavors of Go integer map to BigQuery type INT64. We disallow uint, uint64 and uintptr, because they are (potentially) 64-bit unsigned values, so they cannot be accurately represented as a signed 64-bit integer.
• Go floating-point values map to FLOAT64.
• Go strings map to STRING.
• Go bools map to BOOL.
• time.Time maps to TIMESTAMP.
• Slices and Arrays map to ARRAY<T>.
• Structs and pointers to structs map to STRUCT<...>. We use only the exported fields of the struct, which we can now obtain accurately from the new cloud.google.com/go/internal/fields package.

However, that leaves the three civil time types DATE, TIME and DATETIME, which have no Go equivalent.

We could define these types in Go easily enough. But they are too general to live in this client package. Also, although Date is clearly useful and need for it has come up before, the other two don't seem to be generally missed. Given that the standard library is frozen and golang.org/x is in flux, there doesn't seem to be any great place to put Date.

We also don't really need full-fledged Go types for these BigQuery types, because BigQuery always treats them as strings, on both input and output. (BigQuery Standard SQL provides operations on these types that include componentwise access, but that doesn't concern this client.) We only need Go types for DATE, TIME and DATETIME so that a query parameter can be typed as DATE instead of STRING.

So in the bigquery package we define

type (
    Date string
    Time string
    DateTime string
)

To provide a query parameter of type DATE, one writes Date("2016-11-26").

For consistency, we also use these wrappers when reading fields of these types. (This is not a breaking change; formerly we did not recognize these types and returned an error.)

/cc @mcgreevy @zombiezen @adams-sarah @pongad @nightlyone

[zombiezen]

This seems fine, except I'd like to push back a little on having the three date types be strings. Could we use structs instead?

[jba]

Then we're going down what I called the "full-fledged types" route. Nothing wrong with that per se, but it's a slippery slope: if we have a bigquery.Date type with month, day and year, we'll need methods to convert to/from a string, so we should export those. Maybe also a method to produce a time.Time given a time.Location. Now this is getting to be a useful class. Maybe it shouldn't be in bigquery, but some place more general. But where?

And should we do the same for Time and DateTime? It seems we should, by symmetry. So should we write a civil package for all these civil time concepts?

I'm actually in favor of that, but now we've traveled a long way from bigquery types. So I thought I would keep things as simple as possible.

[derekperkins]

Maybe this kills your mapping, but it seems like it would be easier to use a time�.Time everywhere, just discarding the unnecessary parts based on the underlying type.

I don't have a horse in this race, we use TIMESTAMP everywhere