How to implement Full outer join using MoreLinq?

[gdoron]

Hi,

First, thanks a lot for maintaining this amazingly helpful library!

How can I implement a full outer join two unordered sequences using MoreLinq?
I saw you implemented FullGroupJoin but I'm don't want the grouping stuff.

I'm looking for something like in this SO answer: https://stackoverflow.com/a/13503860/601179 , but with better implementation (using yield return for example)

I've tried OrderedMerge but had troubles using it as the sequences aren't ordered (as I don't care about the order) and don't implement IComparer

var panelAndScrapeCombined = panelDataWithChange.Values.OrderedMerge(
    scrapingRecords,
    panel => new {panel.Domain, panel.RefererFlag, panel.Referer},
    scrape => new {scrape.Domain, scrape.RefererFlag, scrape.Referer},
    panel => new KeywordAnalysisRecord
    {
        Domain = panel.Domain,
        Visits = panel.Visits,
        Change = panel.Change,
        DestUrl = "",
        Position = null,
        PreviousMonthVisits = panel.PreviousMonthVisits,
        SearchEngineFamily = panel.SearchEngineFamily,
        Referer = panel.Referer,
        RefererFlag = RefererFlag
    },
    scrape => new KeywordAnalysisRecord
    {
        Domain = scrape.Domain,
        Visits = 0,
        Change = null,
        DestUrl = scrape.DestUrl,
        Position = scrape.Position,
        PreviousMonthVisits = null,
        SearchEngineFamily = scrape.SearchEngineFamily,
        Referer = scrape.Referer,
        RefererFlag = RefererFlag
    },
    (panel, scrape) => new KeywordAnalysisRecord
    {
        Domain = panel.Domain,
        Visits = panel.Visits,
        Change = panel.Change,
        DestUrl = scrape.DestUrl,
        Position = scrape.Position,
        PreviousMonthVisits = panel.PreviousMonthVisits,
        SearchEngineFamily = panel.SearchEngineFamily,
        Referer = panel.Referer,
        RefererFlag = RefererFlag
    },);

Is there a way of doing that?

Many thanks!

[fsateler]

You can do a full outer join by using FullGroupJoin:

var grouped = panelDataWithChange.Values.FullGroupJoin(
    scrapingRecords, 
    panel => new {panel.Domain, panel.RefererFlag, panel.Referer},
    scrape => new {scrape.Domain, scrape.RefererFlag, scrape.Referer}
    );

var joined = from g in grouped
                from panel in g.First.DefaultIfEmpty()
                from scrape in g.Second.DefaultIfEmpty()
                select new KeywordAnalysisRecord {
                    Domain = g.Key.Domain,
                    Referer = g.Key.Referer,
                    RefererFlag = g.Key.RefererFlag,
                    Visits = panel?.Visits ?? 0,
                    Change = panel?.Change,
                    DestUrl = scrape?.DestUrl ?? "",
                    Position = scrape?.Position,
                    PreviousMonthVisits = panel?.PreviousMonthVisits,
                    SearchEngineFamily = panel?.SearchEngineFamily ?? scrape?.SearchEngineFamily,
                 };

Note the ?. operator because both scrape and panel can be null. I've chosen to simply use ?? to choose between the preferred options but you could use more complicated logic if necessary.

[gdoron]

@fsateler
Thanks for the expedite answer Felipe!
I'm curious, is there an (significant?) overhead to using FullGroupJoin whereas only FullJoin is needed?

Thanks a lot again!

[gdoron]

Also, just as a suggestion, regardless if there's a penalty or not, it would be really nice having a FullOuterJoin that gets a result selector with record and not IEnumerable, even if it's a simple wrapper around FullGroupJoin

IMHO.

[gdoron]

Wait @fsateler, won't your code using FullGroupJoin except from the joining will also do a distinct based on the key?
That's not exactly how FullOuterJoin should behave.
(It's a valid results for my use case as I don't expect to have duplicates, but I guess it's not the case for many others)

[atifaziz]

@gdoron Full outer join is being tracked as #238. An implementation is now proposed in #350 where your feedback and testing would be welcome.