This case, before us for the second time, see 697 F.2d 771 (7th Cir.1983), again raises interesting questions of federal jurisdiction and labor law. The railroad fired Graf in 1977, ostensibly for having falsified his employment application forms--he says really because he had filed a suit against the railroad under the Federal Employers Liability Act, 45 U.S.C. Secs. 51 et seq. The collective bargaining contract between Graf's union and the railroad gave him 60 days in which to appeal his discharge, initially to a higher supervisory layer within the railroad. Graf asked Evans, the chairman of the local union, to file the appeal for him. Evans told him that he would, and later that he had filed it, but after the 60 days had run he discovered it, unfiled, in his pocket. Graf filed suit in state court. Count I of his complaint charged the railroad with having discharged him in retaliation for the FELA suit. Count II charged the union with negligent failure to press his grievance. The union removed the case to federal district court. The wonderfully obscure question of whether the union could remove the entire case by itself, on which see Thomas v. Shelton, 740 F.2d 478, 483-84 (7th Cir.1984), and Bernstein v. Lind-Waldock & Co., 738 F.2d 179, 183 (7th Cir.1984), was made moot by Graf's filing a brand new complaint in federal district court against both defendants after the removal of his state case, see id. at 185-86. The district judge granted the union's motion for summary judgment on Count II and then, on his own initiative, dismissed Count I; so the complaint was dismissed in its entirety.

Graf appealed. We affirmed the dismissal of Count II. Although the Railway Labor Act, 45 U.S.C. Secs. 151 et seq., has been held to give railroad workers an enforceable right to fair representation by their unions in grievance proceedings, we held in our previous opinion (and subsequent cases have reinforced our holding, see, e.g., Camacho v. Ritz-Carlton Water Tower, 786 F.2d 242 (7th Cir.1986), discussing the duty of fair representation under the National Labor Relations Act), that to establish liability the worker must prove more than simple negligence. As the more was missing here, Graf's unfair-representation claim failed. That left his claim of wrongful discharge against the railroad, Count I. We held that dismissal of that count had been proper insofar as the count alleged that the railroad had violated the collective bargaining contract by firing Graf, but not insofar as it alleged a tort under either federal or state law. We therefore remanded Count I, saying, "In the unlikely event that it is found to state a [tort] claim under federal law the district court should retain jurisdiction. But if it is found to state only a state tort claim it should be remanded to the state court where this suit began." 697 F.2d at 781-82. Instead of proceeding thus, the district judge, after a mysterious two-year hiatus following our decision, dismissed Count I on the merits because the only claim it stated was a state law claim preempted by the Railway Labor Act. Graf again appeals, and we must decide whether the district judge was right to terminate the case as he did.

Clearly the judge erred if Count I was not within the jurisdiction of the federal district court, but it may seem obvious that it was within the court's pendent jurisdiction. This is correct but not obvious. Recall that Count II, which presented a federal claim, named only the union as a defendant. If Count I (directed against the railroad) raised no federal claim, it could be brought under the jurisdiction of the district court only by appealing to "pendent party" jurisdiction, an unsettled extension of the more conventional pendent claim jurisdiction. Although we have rejected the use of pendent party jurisdiction to get around the statutory limits on diversity jurisdiction, see Hixon v. Sherwin-Williams Co., 671 F.2d 1005, 1007-09 (7th Cir.1982), the concept may be available where the main claim (in this case the claim against the union) raises a federal question. See, e.g., Zabkowicz v. West Bend Co., 789 F.2d 540, 546-548 (7th Cir.1986); Bernstein v. Lind-Waldock & Co., supra, 738 F.2d at 187. But there is a simpler route to the conclusion that Graf's state law claim against the railroad is within the district court's pendent jurisdiction. Our previous decision interpreted Count I as making a federal claim as well as a state law claim (or maybe as making just a federal claim): namely, a claim of breach of the collective bargaining contract. That claim was not frivolous and it therefore conferred jurisdiction over the state law claim by virtue of pendent claim jurisdiction. So there is no need to rely on pendent party jurisdiction.

Nevertheless the district court's pendent claim jurisdiction does not provide a completely secure basis for the court's action, because of the directive in United Mine Workers v. Gibbs, 383 U.S. 715, 726, 86 S. Ct. 1130, 1139, 16 L. Ed. 2d 218 (1966), that when all federal claims fall out before trial the district court should relinquish its pendent jurisdiction--a course we have repeatedly insisted on. See, e.g., Blau Plumbing, Inc. v. S.O.S. Fix-It, Inc., 781 F.2d 604, 611-12 (7th Cir.1986). Instead of doing that, the district court dismissed Graf's state law claim on the merits. In view of this irregular mode of proceeding we shall consider whether Count I, even when stripped of any claim that the railroad broke the collective bargaining contract (for the previous appeal disposed of that claim), presents a federal question. If so, Count I is within the statutory jurisdiction of the district court and the court was on firm ground in deciding the merits.

Although Count I alleges that Graf was fired for exercising his rights under the Federal Employers Liability Act, it appears that neither that Act nor any other source of federal law creates a federal right against retaliatory discharge. See, e.g., Landfried v. Terminal R.R. Ass'n, 721 F.2d 254 (8th Cir.1983), and cases cited there. On remand the district court so concluded, and Graf does not challenge the conclusion. After this ruling, every explicit federal claim had been excised from the complaint, leaving only--it might appear--a claim for wrongful discharge under the law of Illinois.

But maybe the Railway Labor Act so pervasively occupies the field of railroad labor disputes that a railroad worker's claim of wrongful discharge necessarily invokes federal law. This may seem to be just a fancy way of saying that a state law claim is preempted by federal law, and we know from Franchise Tax Bd. v. Construction Laborers Vacation Trust, 463 U.S. 1, 103 S. Ct. 2841, 77 L. Ed. 2d 420 (1983), that preemption is a defense rather than an ingredient of the claim itself. That was a case of removal, but the principle is the same whenever the issue is whether a claim is within the federal courts' federal-question jurisdiction. To conclude, however, that preemption never affects the nature of the plaintiff's claim would be inconsistent with the rule that a plaintiff cannot by "artful pleading" prevent the removal of a claim that is really federal though the plaintiff tries to conceal its federal nature. See 14A Wright, Miller & Cooper, Federal Practice and Procedure Sec. 3722, at pp. 242-44 (2d ed. 1985).

The question, then, is whether the plaintiff seeks to base his claim on a body of state law that cannot be applied to his case without violating federal law, or on a body of federal law whose provenance he coyly refuses to acknowledge. In the first situation the case is really a state case, blocked by a federal defense; in the second it is a federal case in state wrapping paper. A federal statute could bring about either situation. A statute that merely created a defense to a state claim would bring about the first, so the case to which it applied would not be within the federal-question jurisdiction, which depends on the legal basis of the claim, not the defense. But a statute that took over the whole field, with the result that the claim necessarily arose under federal rather than state law, would place the case within the federal-question jurisdiction. A highly pertinent illustration is suing to enforce a collective bargaining contract that is within the jurisdiction of section 301 of the Taft-Hartley Act, 29 U.S.C. Sec. 185. The Supreme Court has held that section 301 provides the exclusive remedy for such breaches. Therefore anyone who brings suit to redress such a breach is, whether he likes it or not, basing his suit on section 301--on federal law--because federal law, which is supreme, does not allow such a suit to be based on anything else. See Avco Corp. v. Aero Lodge No. 735, Int'l Ass'n of Machinists & Aerospace Workers, 390 U.S. 557, 88 S. Ct. 1235, 20 L. Ed. 2d 126 (1968); Oglesby v. RCA Corp., 752 F.2d 272, 275-78 (7th Cir.1985); Mitchell v. Pepsi-Cola Bottlers, Inc., 772 F.2d 342, 344-45 (7th Cir.1985); Williams v. Caterpillar Tractor Co., 786 F.2d 928 (9th Cir.1986); United Jersey Banks v. Parell, 783 F.2d 360, 365-68 (3d Cir.1986); Comment, Federal Preemption, Removal Jurisdiction, and the Well-Pleaded Complaint Rule, 51 U.Chi.L.Rev. 634 (1984).

The present case would clearly be of the second type if in Illinois, as in some states (see, e.g., Brockmeyer v. Dun & Bradstreet, 113 Wis. 2d 561, 574-75, 335 N.W.2d 834, 841 (1983)), retaliatory discharge were conceived to be the breach of an implied contractual term, here a term read into the collective bargaining contract; for just as section 301 of the Taft-Hartley Act is the exclusive remedy for breaches of such contracts within the scope of that Act, so the Railway Labor Act provides the exclusive remedy for breaches of such contracts within its scope. See Andrews v. Louisville & Nashville R.R., 406 U.S. 320, 92 S. Ct. 1562, 32 L. Ed. 2d 95 (1972). But in Illinois retaliatory discharge is a tort. See, e.g., Koehler v. Illinois Central Gulf R.R., 109 Ill. 2d 473, 94 Ill. Dec. 543, 488 N.E.2d 542 (1985); Midgett v. Sackett-Chicago, Inc., 105 Ill. 2d 143, 85 Ill. Dec. 475, 473 N.E.2d 1280 (1984). Graf can therefore argue that his claim is based on a body of state law; it is not just a dressing up of a federal claim in state law clothing.

But should the federal character of Graf's complaint really turn on whether state law characterizes wrongful discharge as a tort or a breach of contract? Can it be that if wrongful discharge is a tort, the state court decides the issue of preemption subject only to review by the Supreme Court, while if it is a breach of contract, the federal court decides the issue, provided only that the defendant elects to remove the case from state court to federal court, as of course he can do if the case is deemed to arise under federal law? The procedural difference seems too great to allow it to depend on whether state law describes the same conduct--retaliatory discharge--as a tort or as a breach of contract. The state cannot be allowed, merely by the label it attaches to the cause of action, to interfere with the administration of a federal statute. So the remedies provided by federal laws governing disputes over collective bargaining contracts could be exclusive of tort as well as contract remedies. The worker covered by such a contract who brings a tort suit to rectify the consequences of his discharge may be appealing, albeit involuntarily, to federal law, because federal law may provide the sole measure of his rights.

Thus, the simplicity of a test that makes federal jurisdiction depend on whether the complaint is based on state or federal law is illusory in a case where the application of the test depends on first deciding whether federal law merely provides a defense or occupies the field; for deciding whether it is one or the other is not straightforward. Indeed, asking whether federal law provides a defense or occupies the field may just be another way of asking whether the issue of federal preemption shall be decided by a state or a federal court, and perhaps that question should be asked directly, without taking the essentially question-begging step of asking whether the federal statute occupies the field. If the federal statute is deemed merely to create a defense, the state court decides whether it is a good defense; if it is deemed to occupy the field, the federal court decides whether the plaintiff has a cause of action.

If this is what is at stake, the argument for complete preemption may seem weak; the argument amounts to distrust of the ability or willingness of state courts to enforce federal defenses. It would also be simpler to make federal preemption always a defense, so that parties to state court litigation could be sure whether the case was removable to federal court, or if removed whether it could be retained in federal court (the issue in this case), rather than having to speculate about whether the federal court would consider the case one of partial or complete preemption.

Despite these points, there is overwhelming support in the case law for complete preemption in collective bargaining cases; illustrative is our recent decision in Oglesby v. RCA Corp., supra, a case virtually indistinguishable from this one except that it arose under section 301 of the Taft-Hartley Act rather than under the Railway Labor Act. Naturally in an area of such heavy litigation the course of decision does not run entirely true; in particular a very recent Ninth Circuit decision contains an analysis superficially at variance with the approach taken in Oglesby and most other cases. See Williams v. Caterpillar Tractor Co., supra. But that case is far different from the present one. The plaintiffs in Williams had brought suit for breach of contracts allegedly made during a period when they were not part of the bargaining unit and hence were not covered by their employer's collective bargaining contract; as such contracts would not be within the scope of section 301 of the Taft-Hartley Act (or the Railway Labor Act, had that been the applicable statute), the plaintiffs would not have had a federal cuase of action, so there was no issue of the exclusivity of the federal remedy. Where the worker is covered by a collective bargaining contract and therefore has a potential federal remedy, judicial or arbitrable, the cases hold that that remedy is exclusive; the worker has no state remedies. The explanation is the traditional mistrust--a steady theme in federal labor legislation--of state judicial intervention in disputes arising out of collective bargaining activities. But whether soundly based on the history and practicalities of the labor field or not, the principle of complete preemption in collective bargaining matters is too well settled to be disturbed by us; and the force of the principle is no less when the state happens to call wrongful discharge a tort rather than a breach of contract.

Hence Count I, even interpreted to charge that Graf's discharge violated state tort law, was within the district court's statutory jurisdiction; in any event, despite the doubts we expressed earlier, we think that court was entitled to decide the case rather than remand it even if the only basis of federal jurisdiction was pendent jurisdiction. That doctrine is the point of balance between two competing considerations. One is judicial economy: factually related claims between the same parties should be decided in the same court at the same time. The other is state prerogative: a state's own courts should be allowed to decide issues of state law arising in suits between citizens of that state. Normally, if the federal issues fall out of a case before trial, remanding the state issues to the state court will not result in duplicate trials and the anomaly of having a federal court deciding a pure state case becomes the dominant consideration. But like almost all legal generalizations this one has exceptions. Suppose as in this case that a potentially dispositive issue in the pendent claim is an issue of federal law raised by way of defense. Since the federal court is as competent as the state court to decide a question of federal law, why not let the federal court retain the claim, at least for the limited purpose of deciding whether the case should be dismissed on federal grounds? If the federal court decides the question for the defendant and therefore dismisses the case on the merits, the case is over, without requiring either a timeconsuming remand to the state courts or a decision by a federal judge of issues purely of state law. Judicial economy is served, state prerogatives protected.

This is not to suggest that the federal court should decide any and all federal defenses first and the state court should then decide the case in chief and nonfederal defenses. Although it is increasingly common for a trial court to try a potentially dispositive defense first, out of the usual order, in the hope of heading off a lengthy trial, this is more easily done in one court than in two. It may be impossible to do in two, when the basis of the federal court's jurisdiction is pendent jurisdiction. A dismissal based on the district court's relinquishing its pendent jurisdiction is a jurisdictional dismissal, not a final judgment on the merits. It therefore has no preclusive effect, so far as any ruling on the merits that the judge may have made (such as rejecting a federal defense) is concerned. Moreover, it is important that a flexible procedure not be turned into a lockstep. The federal defense may be much more difficult and time-consuming to resolve than the merits of the state law claim; if so the federal court should relinquish jurisdiction over the whole case.

So if it appears either that the federal defense lacks merit or that the underlying state law claim more clearly lacks merit than the federal defense has merit, the district judge normally should relinquish pendent jurisdiction and let the whole case be decided in state court. But if as in this case the district court can tell at a glance that the defendant has a complete and meritorious federal defense, the court has and can exercise the power to retain pendent jurisdiction, adjudicate the defense, and dismiss the case. The Supreme Court said in Gibbs that there might be "situations in which the state claim is so closely tied to questions of federal policy that the argument for exercise of pendent jurisdiction is particularly strong," 383 U.S. at 727, 86 S.Ct. at 1139, and gave as an example where "the allowable scope of the state claim implicates the federal doctrine of pre-emption," id. This observation was made with reference to a case that had gone to trial, but we do not see why it cannot in an appropriate case justify the exercise of pendent jurisdiction even though there has been no trial. Granted, a two-year hiatus between our affirmance of the dismissal of the federal claims and the decision of the federal defense to the pendent claim is not an argument in favor of the procedure.

The exercise of pendent jurisdiction is discretionary, and the discretion runs both ways: discretion (almost unreviewable) to relinquish the jurisdiction, but also discretion to assert it (provided of course that the pendent claim really is the pendant of some colorable federal claim) in appropriate cases even when the federal claims have fallen out before trial. This principle is well recognized in cases where pendent jurisdiction is retained because the statute of limitations would prevent the plaintiff from refiling his pendent claim in state court. See, e.g., Duckworth v. Franzen, 780 F.2d 645, 656 (7th Cir.1985); O'Brien v. Continental Illinois National Bank & Trust Co., 593 F.2d 54, 65 (7th Cir.1979). We see no reason why the district court's discretion cannot sometimes be invoked by defendants, and, as in this case, a pendent claim retained for the purpose of giving it a swift quietus under federal law, as was done in Spielmann v. Anchor Motor Freight, Inc., 551 F. Supp. 817, 824-26 (S.D.N.Y.1982).

These decisions are not evasions of the statement in Gibbs that when the federal claims drop out before trial any pendent state law claims should be dismissed. The point is not that the statement is a dictum; it is; but it has provided the basis for many decisions in this and other circuits, and we have no doubt that it should be treated as authoritative. But there is no indication that the Supreme Court meant to establish an inflexible rule, one that would govern situations remote from the one before the Court in the Gibbs case and that might bring about results opposite to what the Court intended. Judicial economy, the essential policy behind the modern doctrine of pendent jurisdiction which Gibbs created, supports the retention of pendent jurisdiction in any case where substantial judicial resources have already been committed, so that sending the case to another court will cause a substantial duplication of effort. See Rosado v. Wyman, 397 U.S. 397, 405, 90 S. Ct. 1207, 1214, 25 L. Ed. 2d 442 (1970). Trial is simply a convenient benchmark marking the point by which substantial resources have surely been committed. If those resources are expended without a trial, the essential purpose of the doctrine of pendent jurisdiction may be served by retaining the case. See 13B Wright, Miller & Cooper, Federal Practice and Procedure Sec. 3567.1, at p. 136 n. 20 (2d ed.1984); 3A Moore's Federal Practice p 18.07[.1-3], at p. 18-54 n. 40 (2d ed.1985). This case has been in federal court since 1979. The district judge knows the facts and the issues. He should be allowed to resolve the case without making the parties start over in another court system, at least if the case can be resolved now by a ruling on an issue of federal law.

So by either route--federal-question jurisdiction under 28 U.S.C. Sec. 1331 or pendent jurisdiction--the district judge was authorized to decide whether Graf's claim of wrongful discharge is barred by the Railway Labor Act; and we must now consider whether he decided correctly. The Railway Labor Act provides an exclusive remedy, arbitral in nature, for "minor" breaches of collective bargaining contracts in the railroad industry, of which the firing of an individual worker allegedly for cause is a classic illustration. Any worker who tries to litigate such a breach by bringing a suit under state contract law will therefore be met, and defeated, by the defense that the suit is barred by the Act. As an original matter one might have thought that a suit based on a theory of retaliatory discharge did not have its origin in the collective bargaining agreement and was not a breach of contract claim at all, since it would be unaffected if the discharged worker was an employee at will; indeed the principal domain of wrongful discharge is employment at will. But in Jackson v. Consolidated Rail Corp., 717 F.2d 1045 (7th Cir.1983), this court held that the wrongful discharge of a railroad worker can be remedied only by a proceeding under the Railway Labor Act, and in Koehler v. Illinois Central Gulf R.R., supra, the Illinois Supreme Court came to the same conclusion. The best explanation for these decisions is practical: since any retaliatory discharge will violate the collective bargaining agreement, the worker will have a remedy under the Railway Labor Act and it would just complicate labor relations in the railroad industry to give him another remedy under state law. In any event there is much support for the holding in Jackson, see, e.g., Landfried v. Terminal R.R. Ass'n, supra; Minehart v. Louisville & Nashville R.R., 731 F.2d 342 (6th Cir.1984) (per curiam)--and the cases that distinguish Jackson really are distinguishable, see Garibaldi v. Lucky Food Stores, Inc., 726 F.2d 1367, 1371 n. 6 (9th Cir.1984); Gonzalez v. Southern Pac. Transport. Co., 773 F.2d 637, 645 (5th Cir.1985); Gulati v. Burlington Northern R.R., 364 N.W.2d 446 (Minn.App.1985), rev. granted, Minn.S.Ct., May 24, 1985.

Admittedly the principle established by these cases interacts with the narrow scope of the federal tort of unfair representation in an unfortunate way. Through no fault of his, Graf finds himself with no remedy under either the Railway Labor Act or state tort law, the former because Evans, his "griever," was merely negligent (our previous Graf decision), the latter because state law is preempted (Jackson ). It is true that the Railway Labor Act unlike the National Labor Relations Act permits the worker to bypass the union and file his own grievance, but Graf had no reason to think this was necessary--till it was too late. Although Graf never tested the staunchness of the arbitrators' adherence to the 60-day deadline by filing a late claim, arbitrators take deadlines in collective bargaining agreements seriously, see Elkouri & Elkouri, How Arbitration Works 148-49 (3d ed. 1973); Fairweather, Practice and Procedure in Labor Arbitration 101-10 (2d ed.1983), so filing a late claim would probably have been futile.

The result is unsatisfactory, but as is often the case in life the alternatives are not necessarily better. Suits by workers against employers for wrongful discharge, invariably seeking punitive damages and dragging on for years, are not a happy method for resolving disputes over employment; nor are "grievance malpractice" suits against unions in federal court. Graf was fired almost a decade ago and his case has never come to trial. Maybe there is a lesson here. Maybe a better solution than litigation in court is to give workers and their representatives a strong incentive to make the arbitral process set up by the Railway Labor Act work; and maybe the present decision, in any event compelled by precedent, will strengthen that incentive--regrettable though its consequences are for Daniel Graf.

AFFIRMED.

Once again the "forgotten" grievance--not actionable under our cases, see Graf v. Elgin, Joliet & Eastern Railway Co., 697 F.2d 771 (7th Cir.1983); Hoffman v. Lonza, Inc., 658 F.2d 519 (7th Cir.1981); but see 658 F.2d at 523 (Cudahy, J., concurring)--returns to haunt us. Daniel Graf has endured a lengthy, bewildering and utterly fruitless journey through and around the court system--state and federal. He has never had a trial. I certainly cannot quarrel with the majority's hopes for making the arbitral system work. Employees with complaints like Graf's could hardly be worse off than they now seem to be in court. With these reservations and misgivings about the larger problem, I agree with the majority that the district court was authorized to dismiss Count 5 under the authority of Jackson v. Consolidated Rail Corp., 717 F.2d 1045 (7th Cir.1983).1