I hate when that happens, although I guess I'm clear:

Typo in Expected Result: "Fetching should cause duplicate";
should be:"Fetching should NOT cause duplicate";

Typo in EXAMPLE2: "Another used has encountered an issue like"
should be: "Another user has encountered an issue like"

The related source is: https://github.com/php/php-
src/blob/master/ext/sqlite3/sqlite3.c#L1725

The solution to this bug might simply have to be a documentation note indicating 
the SQLite3::exec should be used for inserts instead of SQLite3::query, as 
::query will necessarily re-execute to "step" through.

Before people start recommending a documentation "fix", be aware that I have two 
examples.

The first use case is solved by using exec(). The second is solved by nothing, 
the query is evaluated twice. Those are just two examples demonstrating the same 
issue.

Don't try to fix my examples, try to fix the issue. The culprit seems to be that 
_step is executed for the query once on query(), but then it's executed all over 
again on first fetch, starting at row1 again.

A proposed solution fixing all side effects would be to run step() on query() 
and cache the fetched result, then return it on first fetch, then call step() on 
second fetch etc.:

query(...); // call sqlite3_step, save row1
fetchArray(...); // return saved row1
fetchArray(...); // call sqlite3_step, return row2
fetchArray(...); // call sqlite3_step, return row3
fetchArray(...); // call sqlite3_step, return row4
...

I've been discussing this in #PHP.PECL, with auroraeos and others and basically 
the problem is caused by the OOP interface in the binding.

The binding introduces extra logic so query() in PHP can return a resultset 
object or false etc.

In order to know what to return, query() calls sqlite3_step() which fetches the 
first row of the result.

So far so good. Here's the problem:

The binding then *resets* the query so the first call to fetchArray() returns 
row1 again. This is not library behavior, it's binding behavior that's 
additional logic.

This causes row1 to be computed twice, and queries to run twice etc.

The solution for solving this without introducing BC breaks and interface 
breaks, is for the binding to store the row it fetched during query() with the 
resultset, as outlined in my previous comment, and return it on first call to 
fetchArray, without calling step.

On subsequent calls to fetchArray(), step is called to fetch the other rows.

Either way you look at it, the binding added behavior that isn't in the original 
library, and this is causing performance issues and side effects. The 
responsible thing is to keep the PHP OOP interface compatible, but fix the 
performance issues and side effects.

And what I described is how to do it...

I've run into it trying to do:

    function sq($query, $params = array()) {
        if ($params) {
            $stmt = sdb()->prepare($query);
            if ( ! $stmt) {
                die("sqlite: " . sdb()->lastErrorMsg());
            }
            foreach ($params as $param) {
                $r = call_user_func_array(array($stmt, 'bindValue'), $param);
                if ( ! $r) {
                    die("sqlite: " . sdb()->lastErrorMsg());
                }
            }
            $res = $stmt->execute();
            if ( ! $res) {
                die("sqlite: " . sdb()->lastErrorMsg());
            }
        } else {
            $res = sdb()->query($query);
            if ( ! $res) {
                die("sqlite: " . sdb()->lastErrorMsg());
            }
        }
        $r = array();
        while ($row = $res->fetchArray(SQLITE3_ASSOC)) {
            $r[] = $row;
        }
        return $r;
    }

Made it work by wrapping the last part in `if ($res->numColumns()) {`:

        ...
        }
        if ($res->numColumns()) {
            $r = array();
            while ($row = $res->fetchArray(SQLITE3_ASSOC)) {
                $r[] = $row;
            }
            return $r;
        }
    }

Same problem here, duplicate inserts with "query" method.

Confirmed: <https://3v4l.org/MMGio>. And yes, this is a bug, and
not a documentation problem (nonetheless, DDL and DML statements
should use ::exec()).

I've just encountered this issue on Linux and Version 5.6.30 while using a similar generic wrapper around prepare/bind*/execute(). Is there still intent to fix this after five years?

While the fix for SQLite3::query() is trivial, a general fix for
SQLite3Stmt::execute() is impossible, since it is allowed to
::execute() the same statement multiple times, which might trigger
sqlite3_reset()s at unexpected times (see, for instance, bug
#73530).