Hacker News new | past | comments | ask | show | jobs | submit login
A beginners guide to thinking in SQL (sohamkamani.com)
293 points by craigkerstiens on July 11, 2016 | hide | past | web | favorite | 87 comments

I think this would really benefit from some thoughtful indenting to illustrate the structure. For example, instead of:

  members.firstname AS "First Name",
  members.lastname AS "Last Name"
  FROM borrowings
  JOIN books ON borrowings.bookid=books.bookid
  JOIN members ON members.memberid=borrowings.memberid
  WHERE books.author='Dan Brown';

    members.firstname AS "First Name",
    members.lastname AS "Last Name"
  FROM borrowings
    JOIN books ON borrowings.bookid = books.bookid
    JOIN members ON members.memberid = borrowings.memberid
  WHERE books.author = 'Dan Brown';
This helps reinforce that there is one FROM statement that combines three tables using joins

Some time ago, I settled on a fairly similar idiosyncratic indentation style for SQL:

    members.firstname as "First Name",
    members.lastname as "Last Name"
    books on borrowings.bookid = books.bookid
    members on members.memberid = borrowings.memberid
    books.author = 'Dan Brown';
Normally, I'm a fairly slavish follower of style standards, but almost every bit of SQL I come across in my job is an unreadable mess, even when whoever wrote it did have the vague idea that consistent formatting was good.

Sometimes with complex queries, it can be tough to adhere to a simple set of style rules while also making clear where things like logical units start/stop. Don't get me wrong I'm not disagreeing with you in principle; but acknowledging that the declarative nature of the language cause you to break style for the sake of readability in some cases. I'm not going to post an example of such a case (I'd have to go digging and I'm too tired).

On a different note, I don't like a lot of the styles I do see. I like yours, with the exception of the joins. My convention is a bit different:

      members.firstname as "First Name"
     ,members.lastname as "Last Name"
  FROM borrowings
    JOIN books 
        ON borrowings.bookid = books.bookid
    JOIN members 
        ON members.memberid = borrowings.memberid
  WHERE books.author = 'Dan Brown';
Some of that is old habit (like the comma first thing which I know is out of favor with many) which I found useful when working with long column lists. I'd also typically and systematically alias the tables, but that wasn't done in the examples. Outside of tastes, the only real downside I run into is that it's easy to get too far to the right with the indentation style I use. But it's not that often that I get there.

I haven't had trouble with complicated queries with this style. I've been doing it for ages, though.

There nothing wrong with comma-first, it's sensible; I just never got into the habit. My joins are so I can instantly glance over the tables involved in a query.

(And, oh yes, I, too, normally alias the crap out of tables if there's more than one in a query.)

One of my favorite tools is a SQL auto formatter, http://poorsql.com/ . Also available as an add-in for notepad++, Visual Studio and SSMS.

Fully agree.

Why is it, that SQL generally is so poorly commented? Almost every time I see an SQL script, I cringe for the poor formatting. It's generally not like the writers of these scripts are poor at formatting other code...

SQL has a very large and complex syntax, which makes it hard to come up with a set of formatting rules that fit at least most of the statements. I think the only way to get readable SQL is to use a DSL-based SQL builder for your language - this decouples syntax from meaning and makes good formatting much easier. It also fits much better with your main language.

Maybe a little off topic:

- What's the (hi)story behind SQL putting the SELECT clause first? What are/were the benefits?

- What's the (hi)story behind SQL even differentiating between FROM and JOIN? Isn't it in the end: those are the tables, combine them in this way into one data set?

Any hints? pointers?

SQL uses a natural language syntax.

Regarding your first question - In English, most people would agree that "grab the beer from the fridge" is more natural than "from the fridge grab the beer". For what it's worth I've always considered this aspect of SQL maddening and I always start with SELECT * and work my way back later.

Regarding your second, this is probably because FROM means something different from JOIN. FROM indicates you are starting a block of JOINED tables. Technically if the syntax wasn't natural it might look like:


Given how annoyingly verbose the syntax is already I'm happy to go with what we were given.

Aha! The natural English grammar pointer is a nice one... sure enough I now! remember the "natural language syntax" idea ;)

I'm happy too, no doubt... I guess, given how long it already "survived" SQL is quite "successful". Cheers

You're right on, this is exactly the styling I use with SQL. Indentation makes all the difference in at-a-glance readability.

I have to wonder why a relation algebra language isn't actually algebraic. What about projections in the form of something like c := a x b etc? Nooo, they had to make it look like COBOL.

In databases class, they used Pi for projection, bowtie for natural join, and sigma for row filtration. They added conditions on the operators with subscripts. SQL is just the standardized version of that.

"Standardized"...and without the legibility of symbolic notation and the option to use intermediate named relvars?

It's designed to be "natural", meaning that technical symbols were largely avoided. Someone who knows little about formal notation but a decent amount about databases can usually figure out what's going on. It's also easier to type, seeing how most of the relevant symbols aren't on your average keyboard.

I have seen it taught with operators like that.

Because one of SEQUEL, and thus SQL's design goals was to create a language that's usable by non-programmers of various levels of computer and math fluency. Having to remember a set of symbols just to get a jobs report would be a huge turn-off to a huge subsection of people that the language was intended for.

What's the canonical reference book one should go to regarding SQL? (ie K&R is the book you go to for C)

I understand there are differences between DB vendors, but my question is about general patterns and best practices.

SQL Pocket Guide. Thin, small book. No fluff. Just syntax and solutions for all the major vendors. If I could only have one book while developing it would be this one.


Chris Date's Introduction to SQL [1], or Edgar Codd's last book [2].

Actually both are worth reading to discuss differences in philosophy (e.g. "no NULLs" vs 4-valued logic including "unknown" and "not applicable" NULLs).

[1] https://www.amazon.com/Introduction-SQL-Mastering-Relational...

[2] https://www.amazon.com/Relational-Model-Database-Management-...

First book [1] is not by C. J. Date. So, is this the book you recommend and just mixed up the author's name, or... ?

Good catch. Too late to edit the comment.

Yes, I meant C.J. Date's "Introduction to SQL".

I was trying to get a link to the latest edition which appeared to be the 2nd link on an Amazon search for "Date introduction to SQL"... anyway, it is continuously edited so it's worth seeking the latest version which has twice as many pages as some of the earlier ones.

The thing is, and I probably should have pointed it out in the first comment, sorry about that: I can't find such a book by Cris Date. My guess is you meant "Introduction to Database Systems".

Yes, you are right. Database Systems.

Joe Celko's "SQL for Smarties" ... and then the other SQL books by Celko

Great book! Also, go read jOOQs blog: https://blog.jooq.org/

  > What's the canonical reference book
  > one should go to regarding SQL?
  > (ie K&R is the book you go to for C)
  > I understand there are differences between DB vendors,
  > but my question is about general patterns and best practices.
For 10 years I've looked and never found one. I find Chris Date pedantic. The rest are a mish-mash of cookbooks, vendor-specific ones, and some that try to be the general guide you're looking for but collapse beneath the weight of hundreds of pages of prose that seems to have been written by a committee of robots.

As far as I know, you are probably looking for Jeffrey D. Ullman [1]? I think my lecturer at my local university basically stole [2] when he taught our course :)

Just be aware that he is an academic, so I might compare him more to A.S.Tanenbaum than K&R :-)

[1] http://infolab.stanford.edu/~ullman/ [2] http://infolab.stanford.edu/~ullman/dscb.html#slides

I found that basic courses on relational algebra more useful than most text books. They seem to introduce the mathematics in a more clear concise way than most SQL books.

Not a book, but Stanford's Intro to Databases by Jennifer Widom is offered for free online and I've heard nothing but great things about it.


I found SQL Antipatterns by Bill Karwin to be a great book on the topic of patterns and best practices with SQL. As a bonus, it's relatively platform agnostic if I recall correctly.

Could you write their example

    SELECT members.firstname || ' ' || members.lastname
    AS "Full Name"
    FROM borrowings
    INNER JOIN members
    ON members.memberid=borrowings.memberid
    INNER JOIN books
    ON books.bookid=borrowings.bookid
    WHERE borrowings.bookid IN (SELECT bookid
       FROM books
       WHERE stock>(SELECT avg(stock)
       FROM books))
    GROUP BY members.firstname, members.lastname;
more simply as

    SELECT @avgstock = avg(stock) from books; 

    SELECT members.firstname || ' ' || members.lastname
    AS "Full Name"
    FROM borrowings, members, books
    WHERE members.memberid = borrowings.memberid
      AND books.bookid = borrowings.bookid 
      AND borrowings.bookid IN 
        (SELECT bookid
         FROM books
         WHERE stock > @avgstock))
    GROUP BY members.firstname, members.lastname;
Are those explicit INNER JOIN clauses necessary? By now, most SQL query optimizers can figure out on their own how to join tables.

My impression was that few people use the comma syntax these days, it's old fashioned, harder to read, harder to change, harder for the query plan optimizer, etc.

I know this is an appeal to authority, but the consensus is pretty clear if you search SO and programmer stackexchange:

http://stackoverflow.com/questions/1018822/inner-join-on-vs-... http://programmers.stackexchange.com/questions/78225/using-j...

You're also introducing a lot more concepts by adding variables, and that code wouldn't work on SQL Server, for example, as you need to declare the variable first. Then again the `||` doesn't work on SQL Server either.

As an active DBA I would tell you that most of the time I would expect the comma join syntax to come from someone with grey hair, its an older format that once was standard.

I would also say that if you are suggesting style, you might ommit that INNER from the JOIN (if you are targeting sql server its redundant.)

I tend to mix and match the two. The comma join looks way more natural to me. One thing I do habitually which I don't see other people do much is use aliases.

I learnt SQL through SAS which is where the excessive aliases came from and it kind of stuck with me.

The "comma join" disconnects the join from the condition. It is hard to find the actual join condition between all the "where" conditions. By using JOIN ... ON .., you explicitly link tables with their join condition and also separate join conditions from other filter conditions on the "where".

The "comma join" disconnects the join from the condition.

Yes, because it's more important to first know what's being joined, then traverse that list looking for how it's being joined. You get the full name in the conditionals, so it's not like it's a mystery what they attach to.

But I have grey hair, so there you go.... :)

Comma syntax makes much more sense to me. But now I know that this is not The Way Things Are Done. I also learned from the comments that comma-first isn't recommended. So now my code will be more legible. It's all good.

I'm 45. I'm so old that I'm proficient using (+) to code outer joins in oracle :) Comma-join was the standard for me, and I had a hate-phase towards LEFT/RIGHT JOIN. But I must admin that once I've really tried the "new" join-on syntax it has great advantages that outweighs the pain of change.

Thanks. Good to hear!

I'm 51, and I'm a "full stack" programmer, which used to just be a regular programmer before everybody started overspecializing. When I learned programming, you met a guy, he told you what he wanted, and you made the computer do that. You didn't say "But I'm just the Java guy"

It wasn't that you were an expert in a dozen things, you just figured out whatever the heck you had to do.

I've kept with this philosophy and it's served me well.

So I dive in and out of SQL. I picked up my habits on SQL intensive projects in the 80s and 90s.

It'll be difficult for me to switch until I get into a spot where I'm using it heavily for a week or two. Hopefully I'll remember this thread then and give the new way a shot. Sounds freaky, but hell, I wasn't crazy about pointer arithmetic when I first heard about it either. :)

If you write any decent amount of SQL and you dont alias, I will find you, and I will ... teach you a better way to code.

Most people don't have to do stuff in highly normalized systems, so when you write code that can have dozens of joins and potentially more predicates, you would go insane NOT using aliasing.

> I would expect the comma join syntax to come from someone with grey hair,


(This is like being told by my doctor last week that an injury was "common in older athletes.")

My right knee sends its condolences. :)

> Are those explicit INNER JOIN clauses necessary?

Explicit INNER JOIN is clearer than comma joins, which combine join conditions with filter conditions.

> By now, most SQL query optimizers can figure out on their own how to join tables.

Clarity, rather than efficiency, is the reason to prefer explicit INNER JOINS over comma joins.

Though in this case, the conditions for the explicit INNER JOINS could be simplified by using USING (and, also, eliminating the unnecessary subquery):

    SELECT members.firstname || ' ' || members.lastname
    AS "Full Name"
    FROM borrowings
    INNER JOIN members
    USING memberid
    INNER JOIN books
    USING bookid
    WHERE books.stock > (SELECT avg(stock) FROM books)
    GROUP BY members.firstname, members.lastname;

Thanks for explaining a use of using!

There's so many ways to accomplish the same thing in $SQLEngine that a lot of it comes down to style and clarity/rereadability. I'd prefer keeping the join conditions as USING or ON in most cases because it keeps them close to the statements related to it. Where I work I have to read through a lot of older code that hides them in the where clause and uses 3 letter abbreviations almost exclusively for table aliases which can be pretty tedious to sift through. Again it's mainly down to personal choices though (or in some cases institutional guidelines).

There are dozens of different ways you could write the given query, e.g.,

  Select members.firstname + ' ' + members.lastname as "Full Name"
  From members
  Join borrowings on members.memberid = borrowings.memberid
  Join (
    Select bookid
    from books 
    where stock > (Select avg(stock) from books)
  ) books on borrowings.bookid = books.bookid
Using comma syntax precludes many of these other forms, and therefore I would not recommend teaching it to beginners (or in fact ever using it), as it limits your options.

I was taught the second style of joins in the WHERE clause, but moved to the first style as I saw it was recommended.

Having used both, the INNER JOIN style makes more sense to me, as you are describing the JOINS when you are adding the tables and keeps it separate from the WHERE clause which is really a filtering step.

The second style combines the two and makes and makes it less clear what you are trying to achieve in my opinion.

This is a very naiive and simple question:

I really love the idea of SQL and am looking for a project to use it. It seems like a great way to encapsulate most of the state of your program so you can make your code more decoupled. For crunching logs or other data-dumps it seems like a no brainier, but for everything else I don't really have a a sense for the performance characteristics. Like would it make sense to keep objects' data in an SQL table (like a pImpl) ? is going over the rows of SQL table instead of std::vectors a different order of magnitude performance wise (I assume with something like SQL-Lite I can keep the tables in memory and not on disk).

I guess when is it appropriate to use a database and when is it not?


For context - I'm working almost exclusively in C++ and am usually pretty performance sensitive. In my last job we did radar simulations with lot of entities (radar platforms, aircraft, ballistic missiles etc.) that just lived in STD containers. Decoupling state from the simulation was always a bit of a struggle. Would keeping something like that in an SQL database make sense?

Another example. At the moment I'm working on an art project and I need to keep track of vectors and polygons which will be later used by a ray tracer. I'd also like to keep the state decoupled from my program. Would a SQL database be a bottleneck?

It really depends on your performance characteristics as you mentioned, you can certainly make a database fast as hell but most of the time I would recommend SQL for when you need ACID characteristics(Atomicity, Consistency, Isolation, Durability), not performance per-se because while SQL can be insanely fast, if you do not understand the abstraction it can get slow real fast.

If you find yourself storing large tables of information that is vaguely relational and you value the consistency of your data over raw performance I would generally recommend SQL engines.

I guess i'm more envisioning something like a SQL container. So if you have multiple instances of a pImpl class you would keep the actual state-variables/implementation in a DB. When you want to say do some complicated look up (ex: which objects fit some criteria) you would express that operation in terms of SQL queries

Is there any open-to-everyone SQL database on the net that I can query to try out some simple operations after reading this tutorial?

You can try http://sqlfiddle.com/, although I've not used it.

If you don't object to running one locally, SQLite Manager [1] is a good Firefox addon that lets you visually manage local SQLite databases.

[1] https://addons.mozilla.org/en-US/firefox/addon/sqlite-manage...

Data from Stack Overflow (and other Stack Exchange sites): http://data.stackexchange.com/

Also, you will see many other examples (other people's queries).

Yes, there is: http://sqlfiddle.com/#!15

there's one on your machine called SQLite.

Anyone have a resource for mysql 5.6/5.7 about geoqueries and storing json? Also, resources about how people should build their table schema? It seems like what I've learn from school might be very different on how you should do it industry.

You'll need 5.7 to make full use of JSON features; ditto for geo. MySQL 5.7 is still considered pretty new (GA for ~9 months) so many companies haven't upgraded yet. Amazon RDS and Google Cloud SQL do offer 5.7 though, if you want to try out these features.

Best MySQL resources are generally the manual [0],[1] and Percona's blog [2],[3]

As for MySQL table schema best practices, definitely a broader topic, but yes it tends to differ greatly from what's taught academically. Baron's book [4] is probably the best starting place -- it's a few versions old by this point, but most of the core recommendations around InnoDB tables and indexing in there still apply.

[0] http://dev.mysql.com/doc/refman/5.7/en/spatial-extensions.ht...

[1] https://dev.mysql.com/doc/refman/5.7/en/json.html

[2] https://www.percona.com/blog/2016/02/03/new-gis-features-in-...

[3] https://www.percona.com/blog/2016/03/07/json-document-fast-l...

[4] http://shop.oreilly.com/product/0636920022343.do

Aside from using temp tables is there a way to assign sub-queries as a variable and refer to them later on?

completely simple and contrived for example purposes only, something like:

    active_users = select * from users where user.login > 0;

    select * from active_users where active_users.country is not null;
seems like that would clean up the query nesting quite a bit. Might need some kind of begin / end blocks like with transactions so the query planner would know the scope of those sub-queries.

Yep, a few.

1. Temporary views. This one's my favourite but not everything supports it and sometimes DBMSs have weird limitations, not like allowing parameter binding in them.

2. Temporary tables, as you mention

3. CTEs ("with expressions"). These would be my favourite if they weren't an optimisation fence in postgres.

    WITH active_users
        AS (select * from users where user.login > 0)

    select * from active_users where active_users.country is not null;
4. If you're only using it once, a subselect lets you add a name to it if that's all you're after

    select *
    from (select * from users where user.login > 0) AS active_users
    where active_users.country is not null;
5. If you're using active_users frequently and want it to be globally accessible, a regular view. Often you can make a DB much more useable with a few well placed views like this and you'll need a few less abstractions as well.

Thanks for the options. can you expand on why CTEs are a postgres optimization fence?

as for 4: I am aware of the named subselect but once you add a few of those to the main select statement you start to get into unreadable territory.

back to CTEs, it seems like it should be an easy substitution problem for the postgres query planner. I am not a DB developer so I am probably not aware of all the issues. The CTE seems like the thing I was trying to describe though. Do you think that the optimizations would be more possible with begin and end scopes? It seems like they are part of the same select statement so i doubt scopes would help it.

In practice optimization fence is not normally a problem, and can even be an advantage. The general rule is - just use CTEs. And the specific rule is - learn to profile and optimize using EXPLAIN ANALYZE

Yes, you can do this using CTEs, i.e. WITH clause, pretty much exactly how you suggest.

Or in the case of your simple query, you may also create and use a view or a user defined table valued function, in addition to a CTE. SQL Server also supports table variables in T-SQL, some other dbs may as well.

You can wrap that first part into a INNER JOIN statement or use a CTE like someone else mentioned

Thank you for this. It may be too basic for most of the HN crowd but for me it is a good read for my further development.

Every article that makes it clear that JOIN are effectively operating on tables and thus informally part of the FROM clause (rather than being independent clauses) deserves upvoting.

The one thing I've always wanted to be able to do is to be able to control the projection part of the query. So in other words, I'd love to be able tell the select to display all attributes that are like '%name%'.

Have you tried using CASE with HAVING?

     CASE WHEN attributes LIKE '%name%' THEN name 
          ELSE null END AS name
     FROM table
     HAVING name IS NOT null
Or did you mean selecting by the column's name specifically?

I think GP means "SELECT all columns that have %foo% in the name of the column". It's not possible directly though you can do it with dynamic SQL by querying the data dictionary and dynamically creating the columns retrieved in the SELECT clause.

I think you can search for column names by querying the information_schema to the table schema (at least in postgres)

    column_name ilike '%foo%';
see https://www.postgresql.org/docs/devel/static/infoschema-colu...

For searching for items with '%foo%' in all tables and columns, yeah i think that you're probably right. http://stackoverflow.com/questions/5350088/how-to-search-a-s...

> "For searching for items with '%foo%' in all tables and columns"

To do this you could perform two queries of the information_schema (one query for '%foo%' in tables, one query for '%foo%' in columns) and join them together with a UNION statement (or if you didn't mind duplicates you could use UNION ALL in this case as it's faster and the results from both queries can have the same fields).

I love this guide and will probably steal ideas from it when writing my own SQL tutorials. That said, I don't think SQL is that hard to read compared to other computer languages, at least for non-programmers.

I say this as someone who learned object-oriented programming far before I learned about SQL or its declarative model. SQL was "hard" for me because the declarative style is so functionally different than OOP, and to this day I still haven't ventured to figure out the SQL equivalents of functions, variables, and other mainstays of general programming; I'm satisfied with using a scripting glue language like Python to automate simple SQL SELECT statements.

But I love teaching SQL to non-programmers. You can do a lot of important real-world things with it without having to jump deep into the logical puzzles of "regular" programming. By "deep", I mean things like for-loops; it never fails to surprise me how befuddled beginners are about how for-loops work -- not just the syntax, but the concept of a "block" of code, that for-loops can exist inside of for-loops...or more fundamentally, you can run more than one line of code inside a for-loop.

With SQL, you don't have to worry about any of that. You simply declare what columns you want from a table, what values to filter by, and optionally, how to aggregate it or join it with another table -- concepts that have much clearer analogies to everyday thinking about information (or, at least spreadsheets).

SQL gets complicated when you try to do complicated things. In the OP's opening query, there's a joining of three tables. But without reading it closely, I can already kind of describe the real-world information-need that it fulfills: List the people who have borrowed books that have been borrowed something something to do with `stock`. If I spent a few more minutes, I could figure out what the fuck that subquery, `stock > (SELECT avg(stock) FROM books)` is all about, but the difficulty there is just my ignorance of the domain and/or particular conventions in this hypothetical problem. If I, or even the average novice, wrote this query line-by-line, its purpose would not be so opaque (as writings rarely are to their own authors). The point is that it's much easier for a beginner to author their own SQL without fucking things up in the way you can easily do so with even just a for-loop. I can't imagine that the Python or Ruby equivalent would be nearly as clear as this SQL statement, even for a far simpler kind of join.

My main complaint about SQL is that it seems more logical that the `FROM` clause should come before the `SELECT` statement. Though my initial guess is that that hampers the declarative-clarity of the statement when `JOINS` are involved.

edit: If I could make one suggestion to the submitted article, it would be that the first "Simple Query" is not simple at all:

     SELECT bookid AS "id", title
     FROM books
     WHERE author='Dan Brown';
I'd argue that this is a simpler statement, and one in which the results are immediately recognizable to every spreadsheet user:

     SELECT * FROM books;

The default behavior of Excel is to show all columns of a data file. When you get more advanced as a SQL user (and more advanced in understanding the fundamentals of computation), then you start being selective about SELECT and realizing that, as an end user (nevermind the cost of computation), it's not always ideal to see everything by default:

     SELECT title FROM books;

I would agree except for that once you start adding joins into the mix people get easily confused, while you can certainly express the concept in english (and we often use these sets while thinking) that after teaching 50+ people how to use SQL "well" that visualizing and understanding joins with more than 2 tables that can throw people for loops for years.

Even when correctly writing the syntax to accomplish the goal the underlying connection to everything being a CROSS JOIN with a predicate is not clear to many people who I know who can write SQL effectively.

I agree that JOINS can be complicated, even at their most basic. One very common error is doing a JOIN on what is assumed to be a unique foreign key:

     SELECT * 
     FROM restaurants 
     INNER JOIN inspections
       ON inspections.restaurant_name = restaurants.name
(Ideally, `inspections` would not have `restaurant_name` as a field...but we're talking about legacy government data, in which anything could happen...)

In the best case scenario, the user realizes that the result contains more rows than in `inspections` and, furthermore, realizes that that's an illogical result because (again, ideally) the user realizes that `inspections` has a many-to-one relationship to `restaurants`, and then realizes that `name` shouldn't be used as the foreign key.

The worst case scenario is that the user understands none of those concepts -- the need for foreign keys (in this case) to be unique, what "many-to-one" is -- and also is ignorant of the domain -- e.g. that there may be more than one restaurant with the name "MCDONALDS". And then throw in a few data integrity errors, e.g. `inspections.restaurant_name` contains typos/mistakes/nulls, such that the number of rows in the result set is fewer than number of rows in `inspections`.

But this worst-case user is just satisfied/relieved that the SQL interpreter didn't throw an error and goes on to conduct further erroneous analytical queries on this nonsensical result.

But none of that is the fault of SQL itself. And I have hard time thinking of how exactly SQL, or any programming construct, can make such logical (but not syntactical) errors evident to the novice. At least SQL makes this error fairly evident to subsequent readers of the code.

> And I have hard time thinking of how exactly SQL, or any programming construct, can make such logical (but not syntactical) errors evident to the novice.

SQL could do one (or both) of these to make that easier:

(1) have a preferred syntax for joins that leveraged declared relationship (FK) that required the referenced column(s) of the FK relationship to be a declared candidate key (either PK or subject to a Unique constraint) -- which would make other joins automatically suspect.

EDIT: Incidentally, I've seen proprietary RDBMS languages and visual interfaces (usually, for systems that also supported SQL) that did this exclusively -- you could only join via declared FK relations. This is safer, but probably overly restrictive for general use.

(2) in something otherwise like the existing join syntax, require an explicit modifier to do a join without a constrained-as-unique column on one side.

Ah, I hadn't thought about teaching the novice user to declare their own foreign keys (assuming that the data workflow starts with importing a plaintext CSV into a database) before doing a JOIN. Such a process would force the user to think about what a foreign key column should contain.

(in the past, I've generally avoided teaching table/data altering statements, such as CREATE/DROP/UPDATE, until after thoroughly covering SELECT)

I use SQLite exclusively for teaching, but one thing I miss from PostgreSQL (in my limited usage of it, at least), is how it throws an error when the user attempts to select columns not referenced in a GROUP BY clause:

   SELECT id, last_name, COUNT(1) AS people_with_that_name
   FROM people
   GROUP BY last_name;
`id` is the problematic column here. But in SQLite and MySQL, the interpreter will dutifully include the column even though it has no meaningful interpretation. Come to think of it, SQLite's simplified data types (including implicit type casting) is kind of a pain when it comes to teaching.

What will the id become in that query when you got multiple people with the name? Null?

The solution is very simple. If when you're starting out you don't understand that the 'details' table should be your primary table instead of the 'header' table, then there's a simple trick you can pull... For any inner/left join you create, try altering the query to use a right join and check you don't return more results. That way you can be sure you're targeting the appropriate table.

I'd also recommend using LEFT JOIN over INNER JOIN when starting out exploring data, as the rows with null values can be a useful prompt to check your query logic.

Also, as another SQL tip... always use table aliases. Not only do table aliases make it clear which table you're pulling from, they also help you fill in the right fields if you're using a decent editor that has IntelliSense or something similar.

There's very little reason not to use table aliases, especially if your query is pulling from more than one table.

So the query becomes...

SELECT r., i. FROM restaurants r LEFT JOIN inspections i ON r.name = i.restaurant_name

...and when checking results, just swap LEFT for RIGHT...

SELECT r., i. FROM restaurants r RIGHT JOIN inspections i ON r.name = i.restaurant_name

This last step is unnecessary when you know to target the 'detail' table in the FROM clause.

EDIT: The HN comment system is removing the asterisks after r. and i. . Would just need to add those back in for the queries to work.

I like that idea of teaching LEFT JOINs before INNER JOINs...I've always done the opposite because that's how I learned it (via some book about Microsoft Access) and maybe because INNER JOIN is the default kind of join if you use only "JOIN".

I teach table aliases but only out of necessity when doing nested queries...they aren't hard to grasp but I've assumed it to be less of a priority when students are still struggling with vanilla SELECT/FROM/WHERE. (Also, I learned aliases somewhat late...they didn't seem useful until working in a production environment...just as proper naming conventions in any language seem like trivia until you have to work with someone else).

Glad you like the LEFT JOIN idea. Thank you for being open to new ideas.

Regarding table aliases, I understand the reluctance to overload your students with too much information, but in my experience they really do make writing SQL a lot easier (when coupled with an IntelliSense-capable environment such as SSMS).

Every time a student wants to reference a specific field they type one or two letters (ideally a name that is easy to remember, so if there was an ObjectType table a good alias would be OT, for example), a full stop, then all the available fields from that table are shown in a list. Nothing else is in that list, which makes the decision simpler to make. In SSMS, the IntelliSense usually displays automatically. Sometimes you experience a brief delay from building the cache, but manually toggling IntelliSense usually gets around this (it can be toggled manually with Ctrl + Space, which is an easy shortcut to remember).

On a side note, I tend to use the AS keyword when creating table aliases as I like making them explicit, perhaps that habit will also be beneficial to your students as it's the same keyword used for renaming fields in the SELECT clause also, so in a student's head they can think of AS as rename.

Ironically most programmers I know have no trouble with loops but struggle with SQL. The most common SQL error they make is trying to use loops. (RBAR: Row by Agonizing Row)

As an amusing anecdote, my friend was working with a _really_ slow query (it took literally days to run a query), and their consult/software vendor/whatnot that had written the query kept telling that your data is so large/complex that you really need to look into big data systems like Hadoop.

Turns out the SQL query was running on for-loops. With some tinkering my friend got rid of the loops, and after that the query run in less than three minutes. No more talking about Hadoop there.

If a programmer is struggling with SQL in the way you describe, I'd suggest they learn how to perform MapReduce in the programming language of their choice. That should help give them a better understanding of how to approach SQL. If their language doesn't have good support for the MapReduce approach, a quick introduction in any functional language should help.

I teach SQL to students who may have no programming experience at all, unfortunately

The MapReduce advice was only meant for programmers who had to unlearn imperative programming habits. Non-programmers won't have imperative programming habits to unlearn.

Some of the tables don't work in Safari, but otherwise a good intro.

Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | Legal | Apply to YC | Contact