yavsc/src/Yavsc.Blogs.Tests/BlogsWebServerFixture.cs
Paul Schneider e48ede1e84
acl post: never 500 regression sentinel + async CheckOwner + fixture seed
The hard rule on POST /api/v1/blogacl is: a 500 is never acceptable,
regardless of the payload shape. The prod 500 logged on 2026-08-21 on
mercure was caused by the PostIt client sending { circleId } only, which
the server deserialised into CircleAuthorizationToBlogPost with
BlogPostId = default(long) = 0; EF Core refused the INSERT with
InvalidOperationException: The value of
'CircleAuthorizationToBlogPost.BlogPostId' is unknown. The PostIt fix
lives in b82b6722 (enrich the payload with blogPostId). The server-side
guard lives in this commit:

- BlogAclApiController.CheckOwner is now async and uses FirstOrDefaultAsync
  instead of First, so it does not deadlock the request thread and
  returns false on a missing circle (which the controller already maps
  to ChallengeResult).
- BlogsWebServerFixture now seeds Alice, her Circle and her BlogPost
  in ConfigurePipelineAsync, once at host startup, against the shared
  SqliteConnection (Cache=Shared). EnsureCreated is idempotent and
  runs against the connection that every DbContext resolves through,
  so the test theory can POST payloads with real FK ids against a
  schema that actually has the Circle / BlogSpot tables.
- BlogAclApiTests:
    - PostCircleAuthorization_returns_201_when_payload_mirrors_PostIt_shape
      is the regression sentinel for the prod fix.
    - PostCircleAuthorization_never_returns_500 is a [Theory] over
      several payload shapes; any future commit that reintroduces a
      500 path turns it red. CleanupAcl at the start of each insert-
      bearing test isolates against xUnit's no-guarantee-of-order
      execution: a successful POST in test N would otherwise conflict
      with test N+1 against the same (CircleId, BlogPostId) pair.
2026-08-21 22:00:27 +01:00

383 lines
17 KiB
C#

using Microsoft.AspNetCore.Authorization;
using Microsoft.AspNetCore.Builder;
using Microsoft.Data.Sqlite;
using Microsoft.EntityFrameworkCore;
using Microsoft.Extensions.DependencyInjection;
using Microsoft.IdentityModel.Tokens;
using Yavsc.Blogs.Controllers;
using Yavsc.Models;
using Yavsc.Models.Blog;
using Yavsc.Models.Relationship;
using Yavsc.Services;
using Yavsc.Tests.Shared;
namespace Yavsc.Blogs.Tests;
/// <summary>
/// Shared integration-test host for the Yavsc.Blogs API surface.
/// Specialisation of <see cref="WebHostFixture"/> that wires up
/// only the bits the blog API actually depends on:
///
/// <list type="bullet">
/// <item><description>A SQLite <c>:memory:</c> database
/// (<see cref="Microsoft.EntityFrameworkCore.Sqlite"/>) backed
/// by a single shared <see cref="SqliteConnection"/> held open
/// for the lifetime of the host. SQLite enforces real foreign
/// keys and real transactional semantics, so the tests see the
/// same INSERT-time FK validation a production Postgres host
/// would — unlike the EF Core InMemory provider, which silently
/// ignores FKs and masks bugs that surface only against a real
/// relational engine.</description></item>
/// <item><description>A trivial <see cref="IFileSystemAuthManager"/>
/// stub: the GET index path doesn't read the file system, so any
/// implementation is fine.</description></item>
/// <item><description>The real <c>BlogSpotService</c>, which calls
/// <c>IAuthorizationService.AuthorizeAsync(user, blog, new EditPermission())</c>
/// on PUT. The fixture registers the real
/// <see cref="PermissionHandler"/> so the resource-based ownership
/// check runs end-to-end; tests that want a 204 PUT must sign a
/// JWT whose <c>sub</c> matches the post's <c>AuthorId</c>.</description></item>
/// <item><description>A real <c>AddJwtBearer</c> with HS256,
/// sharing its <see cref="TestTokenIssuer.SigningKey"/> with the
/// token issuer. The production OIDC discovery path is bypassed:
/// the test host validates tokens locally, against the static
/// signing key, so no IdP is required to exercise auth.</description></item>
/// <item><description>The production <c>BlogScope</c> policy
/// (RequireAuthenticatedUser + RequireClaim("scope", "blogs"))
/// registered verbatim. Tests that omit the bearer header exercise
/// the unauthenticated path and get 401.</description></item>
/// </list>
///
/// No IdentityServer, no SMTP, no static assets — the Org fixture
/// owns all of that and we don't need any of it for blog integration
/// tests. Marked <see cref="CollectionDefinitionAttribute"/> so the
/// host is shared across every <c>[Collection("Yavsc Blogs")]</c>
/// test class: one host, one SQLite DB, one Kestrel port.
/// </summary>
[CollectionDefinition("Yavsc Blogs")]
public sealed class BlogsWebServerFixture : WebHostFixture
{
protected override int HttpsPort => 5103;
public long CircleId { get; private set; }
public long PostId { get; private set; }
// A single SqliteConnection held open at the static level,
// mirroring how Yavsc.Org.Tests.WebServerFixture hoists its
// shared configuration into static slots. Closing the
// connection destroys the in-memory database — so we close
// it only when the last fixture instance is disposed (see
// Dispose below), exactly when WebHostFixture tears down the
// host.
private static SqliteConnection? _sharedSqliteConnection;
private static readonly object _sqliteLock = new();
protected override WebApplication BuildApp(WebApplicationBuilder builder)
{
// Open the shared in-memory connection lazily on the first
// fixture construction. Subsequent constructions (xUnit
// creates one fixture instance per IClassFixture) reuse
// the same connection so all DbContexts across all tests
// see the same database.
SqliteConnection sharedConnection;
lock (_sqliteLock)
{
if (_sharedSqliteConnection is null)
{
// Mode=Memory + Cache=Shared gives us a named
// in-memory database that every connection string
// referencing "File:YavscBlogsTests?mode=memory&cache=shared"
// will resolve to the same backing store, as long
// as at least one SqliteConnection stays open
// against it.
_sharedSqliteConnection = new SqliteConnection(
"Data Source=YavscBlogsTests;Mode=Memory;Cache=Shared");
_sharedSqliteConnection.Open();
}
sharedConnection = _sharedSqliteConnection;
}
builder.Services.AddDbContext<ApplicationDbContext>(opt =>
// UseSqlite(DbConnection) keeps the connection we just
// opened alive for the DbContext's lifetime, instead of
// letting EF open and close its own. Without this,
// each DbContext would get a fresh connection pointing
// at an empty :memory: store and nothing would persist
// across requests.
opt.UseSqlite(sharedConnection));
// Trivial file-system auth: the GET index path never calls
// into it, but the DI container needs an instance.
builder.Services.AddSingleton<IFileSystemAuthManager>(
new NoopFileSystemAuthManager());
// Real BlogSpotService — same instance the production host
// builds (ApplicationDbContext, IAuthorizationService,
// IFileSystemAuthManager). With PermissionHandler registered
// below, Modify() now answers "is the caller the author of
// the post?" for real, which is exactly what we want to
// assert in the PUT tests.
builder.Services.AddScoped<BlogSpotService>();
// The real PermissionHandler: BlogSpotService calls
// IAuthorizationService.AuthorizeAsync(user, blog, new
// EditPermission()) on Modify, and PermissionHandler
// resolves it via IsOwner(user, blog) — i.e. blog.AuthorId
// == user.GetUserId(). To PUT a post, the test JWT must
// carry sub == post.AuthorId.
builder.Services.AddScoped<IAuthorizationHandler, PermissionHandler>();
// The BlogApiController is reached through MVC. AddControllers()
// by default scans the test assembly only; we explicitly add the
// Yavsc.Blogs application part so the controller is discovered
// and routed.
builder.Services.AddControllers()
.AddApplicationPart(typeof(BlogApiController).Assembly);
// Production BlogScope policy, verbatim. Two requirements:
// 1. RequireAuthenticatedUser: a request with no bearer
// token (or an invalid one) will be rejected.
// 2. RequireClaim("scope", "blogs"): the JWT must carry a
// "scope" claim whose value is "blogs".
// TestTokenIssuer.Issue() defaults to scope=blogs; the
// GetBlog_returns_401_when_no_token test omits the token
// entirely and asserts the policy fails closed.
builder.Services.AddAuthorization(opt =>
{
opt.AddPolicy("BlogScope", policy =>
{
policy.RequireAuthenticatedUser()
.RequireClaim("scope", "blogs");
});
});
// Real JWT Bearer authentication, sharing the signing key
// with TestTokenIssuer. No Authority → no OIDC discovery,
// no IdP roundtrip; the middleware validates the signature
// and the standard claims against the static configuration
// below. Production uses AddYavscJwtBearer with an IdP, but
// for the unit-test host that path is unwanted coupling.
builder.Services.AddAuthentication("Bearer")
.AddJwtBearer("Bearer", options =>
{
options.IncludeErrorDetails = true;
// MapInboundClaims = false here mirrors the
// JwtSecurityTokenHandler.DefaultInboundClaimTypeMap
// .Clear() in TestTokenIssuer: the validation
// pipeline must not rewrite "sub" to
// ClaimTypes.NameIdentifier, otherwise the
// PermissionHandler ownership check sees a null
// user id and rejects every PUT.
options.MapInboundClaims = false;
options.TokenValidationParameters = new TokenValidationParameters
{
ValidateIssuer = true,
ValidIssuer = TestTokenIssuer.Issuer,
ValidateAudience = false,
ValidateLifetime = true,
ValidateIssuerSigningKey = true,
IssuerSigningKey = TestTokenIssuer.SigningKey,
// "sub" stays "sub" (MapInboundClaims only
// remaps long Microsoft claim URIs, not sub).
// UserHelpers.GetUserId reads sub directly.
NameClaimType = "sub",
RoleClaimType = Yavsc.Constants.RoleClaimType,
};
});
return builder.Build();
}
protected override async Task<WebApplication> ConfigurePipelineAsync(WebApplication app)
{
// UseDeveloperExceptionPage gives full stack traces on
// 500s during tests — much easier to debug than the
// default empty InternalServerError body. Production
// (Yavsc.Org) wires its own exception handler; this
// fixture is test-only.
app.UseDeveloperExceptionPage();
app.UseRouting();
app.UseAuthentication();
app.UseAuthorization();
app.MapControllers();
// EnsureCreated + seed alice, run once at host startup.
// EnsureCreated is idempotent (creates only the tables that
// don't exist yet) and runs against the shared
// SqliteConnection (Cache=Shared), so every DbContext that
// resolves through this fixture's host sees the same schema.
// We do NOT call EnsureDeleted: the SqliteConnection is held
// open at the static level and closing it destroys the
// :memory: store for every other DbContext — the org
// fixture can afford EnsureDeleted because its store is
// built fresh per fixture, but the blogs fixture's static
// connection outlives a single fixture instance.
using (var seedScope = app.Services.CreateScope())
{
var db = seedScope.ServiceProvider
.GetRequiredService<ApplicationDbContext>();
db.Database.EnsureCreated();
if (!db.Users.Any(u => u.Id == "alice"))
{
db.Users.Add(new ApplicationUser
{
Id = "alice",
UserName = "alice",
Email = "alice@example.com",
EmailConfirmed = true,
FullName = "Alice Dupont",
Avatar = "/avatars/alice.png",
});
db.SaveChanges();
// Inline the seed of the circle + post. We don't
// call SeedCircle/SeedBlogPost (the instance helpers)
// because those resolve through this.Services, which
// is null until WebHostFixture.InitializeAsync has
// finished wiring the shared slot — i.e. after this
// method returns. Use app.Services directly.
var circle = new Circle
{
OwnerId = "alice",
Name = "test",
Public = true,
};
db.Circle.Add(circle);
db.SaveChanges();
CircleId = circle.Id;
var post = new BlogPost
{
AuthorId = "alice",
Title = "Billet ACL test",
Article = "Test article body.",
DateCreated = DateTime.UtcNow,
DateModified = DateTime.UtcNow,
};
db.BlogSpot.Add(post);
db.SaveChanges();
PostId = post.Id;
}
}
await Task.CompletedTask;
return app;
}
public override void Dispose()
{
try
{
base.Dispose();
}
finally
{
// Close the shared SQLite connection only when the
// last fixture instance goes away, matching the
// lifetime contract of WebHostFixture.Dispose. We
// rely on base.Dispose's _instanceCount decrement
// having run, so we close only if the host is gone
// (base already nulled _app when count==0).
lock (_sqliteLock)
{
if (_sharedSqliteConnection is not null)
{
// Synchronous close: SQLite's Close() is
// documented as safe to call from a sync
// context and avoids the GetAwaiter().GetResult()
// pattern that's historically caused teardown
// hangs in this repo's async pipeline.
_sharedSqliteConnection.Close();
_sharedSqliteConnection.Dispose();
_sharedSqliteConnection = null;
}
}
}
}
/// <summary>Seed an <see cref="ApplicationUser"/> in the shared
/// SQLite store, so tests that POST/PUT/DELETE a
/// <c>BlogPost</c> (whose <c>AuthorId</c> is a FK to
/// <c>AspNetUsers.Id</c>) don't trip the FK constraint that
/// SQLite enforces but the EF Core InMemory provider silently
/// ignored. Idempotent on <paramref name="userName"/>: a
/// second call for the same id is a no-op (the user already
/// exists).</summary>
/// <param name="userName">Both the PK id and the login name.
/// The JWT subject in tests is this same string, so seeding
/// this id is enough to make the FK from a
/// <c>BlogPost.AuthorId</c> resolve.</param>
/// <param name="configure">Optional hook to fill in fields
/// like <c>FullName</c> / <c>Avatar</c> / <c>EmailConfirmed</c>
/// that downstream tests assert on.</param>
public ApplicationUser SeedUser(string userName, Action<ApplicationUser>? configure = null)
{
using var scope = Services.CreateScope();
var db = scope.ServiceProvider.GetRequiredService<ApplicationDbContext>();
var existing = db.Users.SingleOrDefault(u => u.Id == userName);
if (existing != null) return existing;
// Email is an alternate key on ApplicationUser; seeding
// it explicitly avoids the InMemory provider's null-claim
// tracking quirk (cf. PublishEndpointTests.ResetDatabase)
// and keeps the column shape realistic for prod.
var user = new ApplicationUser
{
Id = userName,
UserName = userName,
Email = $"{userName}@example.test",
};
configure?.Invoke(user);
db.Users.Add(user);
db.SaveChanges();
return user;
}
/// <summary>Trivial <see cref="IFileSystemAuthManager"/> stub. The
/// blog API endpoints exercised by the first tests don't read the
/// file system, so the implementation can be a no-op.</summary>
private sealed class NoopFileSystemAuthManager : IFileSystemAuthManager
{
public FileAccessRight GetFilePathAccess(System.Security.Claims.ClaimsPrincipal user, string fileRelativePath)
=> FileAccessRight.None;
public void SetAccess(long circleId, string normalizedFullPath, FileAccessRight access)
{
}
}
/// <summary>Create a circle owned by <paramref name="ownerId"/>
/// directly in the SQLite store and return its server-assigned
/// id.</summary>
private long SeedCircle(string ownerId, string name, bool isPublic = false)
{
using var scope = Services.CreateScope();
var db = scope.ServiceProvider.GetRequiredService<ApplicationDbContext>();
var circle = new Circle { OwnerId = ownerId, Name = name, Public = isPublic };
db.Circle.Add(circle);
db.SaveChanges();
return circle.Id;
}
/// <summary>Create a blog post owned by <paramref name="authorId"/>
/// directly in the SQLite store and return its server-assigned
/// id.</summary>
private long SeedBlogPost(string authorId, string title)
{
using var scope = Services.CreateScope();
var db = scope.ServiceProvider.GetRequiredService<ApplicationDbContext>();
var post = new BlogPost
{
AuthorId = authorId,
Title = title,
Article = "Test article body.",
DateCreated = DateTime.UtcNow,
DateModified = DateTime.UtcNow,
};
db.BlogSpot.Add(post);
db.SaveChanges();
return post.Id;
}
}