add S3 integration test/update readme

README.md

@@ -2,7 +2,7 @@
 
 [![Build Status](https://drone.lerch.org/api/badges/lobo/aws-sdk-for-zig/status.svg?ref=refs/heads/master)](https://drone.lerch.org/api/badges/lobo/aws-sdk-for-zig/)
 
-This SDK currently supports all AWS services except S3. See TODO list below.
+This SDK currently supports all AWS services. S3 has basic support
 
 Current executable size for the demo is 1.7M (90k of which is the AWS PEM file,
 and approximately 600K for XML services) after compiling with -Drelease-safe and
@@ -42,15 +42,13 @@ for posterity, and supports x86_64 linux. The old branch is deprecated.
 
 ## Limitations
 
-There are many nuances of AWS V4 signature calculation. S3 is not supported
-because it uses many of these edge cases. Also endpoint calculation is special
-for S3. WebIdentityToken is not yet implemented.
+There are many nuances of AWS V4 signature calculation, and not all edge cases
+of S3 are handled. WebIdentityToken is not yet implemented.
 
 TODO List:
 
-* Implement initial S3 support. This involves:
-  * Implementation of AWS SigV4 signature calculation for S3, which is unique
-  * Implementation of S3 endpoint calculation, which is also unique to this service
+* Implement more robust S3 support. Keys with slashes in the name are currently
+  causing a SignatureDoesNotMatch error
 * Bump to zig 0.9.1. iguanaTLS, used in zFetch is still [working out 0.9.1 issues](https://github.com/alexnask/iguanaTLS/pull/29)
 * Implement sigv4a signing
 * Implement jitter/exponential backoff

src/aws.zig

@@ -4,6 +4,7 @@ const awshttp = @import("aws_http.zig");
 const json = @import("json.zig");
 const url = @import("url.zig");
 const case = @import("case.zig");
+const date = @import("date.zig");
 const servicemodel = @import("servicemodel.zig");
 const xml_shaper = @import("xml_shaper.zig");
 
@@ -126,7 +127,7 @@ pub fn Request(comptime action: anytype) type {
                     // query string. TODO: RTFM on zig to figure out why
                     aws_request.query = try options.client.allocator.dupe(u8, aws_request.path[inx..]);
                     aws_request.path = try options.client.allocator.dupe(u8, aws_request.path[0..inx]);
-                    log.debug("inx: {d}\n\tnew path: {s}\n\tnew query: {s}", .{ inx, aws_request.path, aws_request.query });
+                    // log.debug("inx: {d}\n\tnew path: {s}\n\tnew query: {s}", .{ inx, aws_request.path, aws_request.query });
                     options.client.allocator.free(orig_path);
                     options.client.allocator.free(orig_query);
                 }
@@ -263,31 +264,78 @@ pub fn Request(comptime action: anytype) type {
                 return error.HttpFailure;
             }
 
-            var fullResponse = try getFullResponseFromBody(aws_request, response, options);
+            var full_response = try getFullResponseFromBody(aws_request, response, options);
+            errdefer full_response.deinit();
+
             // Fill in any fields that require a header. Note doing it post-facto
             // assumes all response header fields are optional, which may be incorrect
             if (@hasDecl(action.Response, "http_header")) {
-                inline for (std.meta.fields(@TypeOf(action.Response.http_header))) |f| {
-                    const header_name = @field(action.Response.http_header, f.name);
-                    for (response.headers) |h| {
-                        if (std.ascii.eqlIgnoreCase(h.name, header_name)) {
-                            log.debug("Response header {s} configured for field. Setting {s} = {s}", .{ h.name, f.name, h.value });
-                            const field_type = @TypeOf(@field(fullResponse.response, f.name));
-                            // TODO: Fix this. We need to make this much more robust
-                            // The deal is we have to do the dupe though
-                            // Also, this is a memory leak atm
-                            if (field_type == ?[]const u8) {
-                                @field(fullResponse.response, f.name) = try options.client.allocator.dupe(u8, (try coerceFromString(field_type, h.value)).?);
-                            } else {
-                                @field(fullResponse.response, f.name) = try coerceFromString(field_type, h.value);
-                            }
+                log.debug("Checking headers based on type: {s}", .{@typeName(action.Response)});
+                const HeaderInfo = struct {
+                    name: []const u8,
+                    T: type,
+                    header_name: []const u8,
+                };
+                comptime var fields = [_]?HeaderInfo{null} ** std.meta.fields(@TypeOf(action.Response.http_header)).len;
+                inline for (std.meta.fields(@TypeOf(action.Response.http_header))) |f, inx| {
+                    fields[inx] = HeaderInfo{
+                        .name = f.name,
+                        .T = @TypeOf(@field(full_response.response, f.name)),
+                        .header_name = @field(action.Response.http_header, f.name),
+                    };
+                }
+                inline for (fields) |f| {
+                    for (response.headers) |header| {
+                        if (std.mem.eql(u8, header.name, f.?.header_name)) {
+                            log.debug("Response header {s} configured for field. Setting {s} = {s}", .{ header.name, f.?.name, header.value });
+                            // TODO: Revisit return for this function. At the moment, there
+                            // is something in the compiler that is causing the inline for
+                            // surrounding this to start repeating elements
+                            //
+                            // https://github.com/ziglang/zig/issues/10507
+                            //
+                            // This bug is also relevant to some of the many,
+                            // many different methods used to try to work around:
+                            // https://github.com/ziglang/zig/issues/10029
+                            //
+                            // Note: issues found on zig 0.9.0
+                            setHeaderValue(
+                                options.client.allocator,
+                                &full_response.response,
+                                f.?.name,
+                                f.?.T,
+                                header.value,
+                            ) catch |e| {
+                                log.err("Could not set header value: Response header {s}. Field {s}. Value {s}", .{ header.name, f.?.name, header.value });
+                                log.err("Error: {}", .{e});
+                                if (@errorReturnTrace()) |trace| {
+                                    std.debug.dumpStackTrace(trace.*);
+                                }
+                            };
 
                             break;
                         }
                     }
                 }
             }
-            return fullResponse;
+            return full_response;
+        }
+
+        fn setHeaderValue(
+            allocator: std.mem.Allocator,
+            response: anytype,
+            comptime field_name: []const u8,
+            comptime field_type: type,
+            value: []const u8,
+        ) !void {
+            // TODO: Fix this. We need to make this much more robust
+            // The deal is we have to do the dupe though
+            // Also, this is a memory leak atm
+            if (field_type == ?[]const u8) {
+                @field(response, field_name) = try allocator.dupe(u8, value);
+            } else {
+                @field(response, field_name) = try coerceFromString(field_type, value);
+            }
         }
 
         fn getFullResponseFromBody(aws_request: awshttp.HttpRequest, response: awshttp.HttpResult, options: Options) !FullResponseType {
@@ -297,6 +345,26 @@ pub fn Request(comptime action: anytype) type {
             var expected_body_field_len = std.meta.fields(action.Response).len;
             if (@hasDecl(action.Response, "http_header"))
                 expected_body_field_len -= std.meta.fields(@TypeOf(action.Response.http_header)).len;
+            if (@hasDecl(action.Response, "http_payload")) {
+                var rc = FullResponseType{
+                    .response = .{},
+                    .response_metadata = .{
+                        .request_id = try requestIdFromHeaders(aws_request, response, options),
+                    },
+                    .parser_options = .{ .json = .{} },
+                    .raw_parsed = .{ .raw = .{} },
+                    .allocator = options.client.allocator,
+                };
+                var body_field = @field(rc.response, action.Response.http_payload);
+                const BodyField = @TypeOf(body_field);
+                if (BodyField == []const u8 or BodyField == ?[]const u8) {
+                    expected_body_field_len = 0;
+                    // We can't use body_field for this set - only @field will work
+                    @field(rc.response, action.Response.http_payload) = try options.client.allocator.dupe(u8, response.body);
+                    return rc;
+                }
+                rc.deinit();
+            }
 
             // We don't care about the body if there are no fields we expect there...
             if (std.meta.fields(action.Response).len == 0 or expected_body_field_len == 0) {
@@ -568,15 +636,31 @@ pub fn Request(comptime action: anytype) type {
     };
 }
 
-fn coerceFromString(comptime T: type, val: []const u8) !T {
+fn coerceFromString(comptime T: type, val: []const u8) anyerror!T {
     if (@typeInfo(T) == .Optional) return try coerceFromString(@typeInfo(T).Optional.child, val);
     // TODO: This is terrible...fix it
     switch (T) {
         bool => return std.ascii.eqlIgnoreCase(val, "true"),
-        i64 => return try std.fmt.parseInt(T, val, 10),
+        i64 => return parseInt(T, val) catch |e| {
+            log.err("Invalid string representing i64: {s}", .{val});
+            return e;
+        },
         else => return val,
     }
 }
+fn parseInt(comptime T: type, val: []const u8) !T {
+    const rc = std.fmt.parseInt(T, val, 10);
+    if (!std.meta.isError(rc)) return rc;
+
+    if (T == i64) {
+        return date.parseEnglishToTimestamp(val) catch |e| {
+            log.err("Error coercing date string '{s}' to timestamp value", .{val});
+            return e;
+        };
+    }
+    log.err("Error parsing string '{s}' to integer", .{val});
+    return rc;
+}
 
 fn generalAllocPrint(allocator: std.mem.Allocator, val: anytype) !?[]const u8 {
     switch (@typeInfo(@TypeOf(val))) {
@@ -781,6 +865,17 @@ fn FullResponse(comptime action: anytype) type {
                     }
                 }
             }
+            if (@hasDecl(Response, "http_payload")) {
+                var body_field = @field(self.response, Response.http_payload);
+                const BodyField = @TypeOf(body_field);
+                if (BodyField == []const u8) {
+                    self.allocator.free(body_field);
+                }
+                if (BodyField == ?[]const u8) {
+                    if (body_field) |f|
+                        self.allocator.free(f);
+                }
+            }
         }
     };
 }

src/date.zig

@@ -57,6 +57,86 @@ pub fn timestampToDateTime(timestamp: i64) DateTime {
     return DateTime{ .day = day, .month = month, .year = year, .hour = hours, .minute = minutes, .second = seconds };
 }
 
+pub fn parseEnglishToTimestamp(data: []const u8) !i64 {
+    return try dateTimeToTimestamp(try parseEnglishToDateTime(data));
+}
+
+const EnglishParsingState = enum { Start, Day, Month, Year, Hour, Minute, Second, End };
+/// Converts a string to a timestamp value. May not handle dates before the
+/// epoch. Dates should look like "Fri, 03 Jun 2022 18:12:36 GMT"
+pub fn parseEnglishToDateTime(data: []const u8) !DateTime {
+    // Fri, 03 Jun 2022 18:12:36 GMT
+    if (!std.mem.endsWith(u8, data, "GMT")) return error.InvalidFormat;
+
+    var start: usize = 0;
+    var state = EnglishParsingState.Start;
+    // Anything not explicitly set by our string would be 0
+    var rc = DateTime{ .year = 0, .month = 0, .day = 0, .hour = 0, .minute = 0, .second = 0 };
+    for (data) |ch, i| {
+        _ = i;
+        switch (ch) {
+            ',' => {},
+            ' ', ':' => {
+                // State transition
+
+                // We're going to coerce and this might not go well, but we
+                // want the compiler to create checks, so we'll turn on
+                // runtime safety for this block, forcing checks in ReleaseSafe
+                // ReleaseFast modes.
+                const next_state = try endEnglishState(state, &rc, data[start..i]);
+                state = next_state;
+                start = i + 1;
+            },
+            else => {}, // We need to be pretty trusting on this format...
+        }
+    }
+    return rc;
+}
+
+fn endEnglishState(current_state: EnglishParsingState, date: *DateTime, prev_data: []const u8) !EnglishParsingState {
+    var next_state: EnglishParsingState = undefined;
+    log.debug("endEnglishState. Current state '{s}', data: {s}", .{ current_state, prev_data });
+
+    // Using two switches is slightly less efficient, but more readable
+    switch (current_state) {
+        .End => return error.IllegalStateTransition,
+        .Start => next_state = .Day,
+        .Day => next_state = .Month,
+        .Month => next_state = .Year,
+        .Year => next_state = .Hour,
+        .Hour => next_state = .Minute,
+        .Minute => next_state = .Second,
+        .Second => next_state = .End,
+    }
+
+    switch (current_state) {
+        .Year => date.year = try std.fmt.parseUnsigned(u16, prev_data, 10),
+        .Month => date.month = try parseEnglishMonth(prev_data),
+        .Day => date.day = try std.fmt.parseUnsigned(u8, prev_data, 10),
+        .Hour => date.hour = try std.fmt.parseUnsigned(u8, prev_data, 10),
+        .Minute => date.minute = try std.fmt.parseUnsigned(u8, prev_data, 10),
+        .Second => date.second = try std.fmt.parseUnsigned(u8, prev_data, 10),
+        .Start => {},
+        .End => return error.InvalidState,
+    }
+    return next_state;
+}
+
+fn parseEnglishMonth(data: []const u8) !u8 {
+    if (std.ascii.startsWithIgnoreCase(data, "Jan")) return 1;
+    if (std.ascii.startsWithIgnoreCase(data, "Feb")) return 2;
+    if (std.ascii.startsWithIgnoreCase(data, "Mar")) return 3;
+    if (std.ascii.startsWithIgnoreCase(data, "Apr")) return 4;
+    if (std.ascii.startsWithIgnoreCase(data, "May")) return 5;
+    if (std.ascii.startsWithIgnoreCase(data, "Jun")) return 6;
+    if (std.ascii.startsWithIgnoreCase(data, "Jul")) return 7;
+    if (std.ascii.startsWithIgnoreCase(data, "Aug")) return 8;
+    if (std.ascii.startsWithIgnoreCase(data, "Sep")) return 9;
+    if (std.ascii.startsWithIgnoreCase(data, "Oct")) return 10;
+    if (std.ascii.startsWithIgnoreCase(data, "Nov")) return 11;
+    if (std.ascii.startsWithIgnoreCase(data, "Dec")) return 12;
+    return error.InvalidMonth;
+}
 pub fn parseIso8601ToTimestamp(data: []const u8) !i64 {
     return try dateTimeToTimestamp(try parseIso8601ToDateTime(data));
 }
@@ -327,3 +407,8 @@ test "Convert ISO8601 string to timestamp" {
 test "Convert datetime to timestamp before 1970" {
     try std.testing.expectEqual(@as(i64, -449392815), try dateTimeToTimestamp(DateTime{ .year = 1955, .month = 10, .day = 05, .hour = 16, .minute = 39, .second = 45 }));
 }
+
+test "Convert whatever AWS is sending us to timestamp" {
+    const string_date = "Fri, 03 Jun 2022 18:12:36 GMT";
+    try std.testing.expectEqual(DateTime{ .year = 2022, .month = 06, .day = 03, .hour = 18, .minute = 12, .second = 36 }, try parseEnglishToDateTime(string_date));
+}

src/main.zig

@@ -12,19 +12,22 @@ pub fn log(
     args: anytype,
 ) void {
     // Ignore aws_signing messages
-    if (verbose < 2 and scope == .aws_signing and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
+    if (verbose < 3 and scope == .aws_signing and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
         return;
     // Ignore aws_credentials messages
-    if (verbose < 2 and scope == .aws_credentials and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
+    if (verbose < 3 and scope == .aws_credentials and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
         return;
     // Ignore xml_shaper messages
-    if (verbose < 2 and scope == .xml_shaper and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
+    if (verbose < 3 and scope == .xml_shaper and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
         return;
     // Ignore date messages
-    if (verbose < 2 and scope == .date and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
+    if (verbose < 3 and scope == .date and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
         return;
     // Ignore awshttp messages
-    if (verbose < 1 and scope == .awshttp and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
+    if (verbose < 2 and scope == .awshttp and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
+        return;
+
+    if (verbose < 1 and scope == .aws and @enumToInt(level) >= @enumToInt(std.log.Level.debug))
         return;
     const scope_prefix = "(" ++ @tagName(scope) ++ "): ";
     const prefix = "[" ++ @tagName(level) ++ "] " ++ scope_prefix;
@@ -50,7 +53,7 @@ const Tests = enum {
     rest_json_1_work_with_lambda,
     rest_xml_no_input,
     rest_xml_anything_but_s3,
-    // rest_xml_work_with_s3,
+    rest_xml_work_with_s3,
 };
 
 pub fn main() anyerror!void {
@@ -231,6 +234,77 @@ pub fn main() anyerror!void {
                 std.log.info("key group list max: {d}", .{list.max_items});
                 std.log.info("key group quantity: {d}", .{list.quantity});
             },
+            .rest_xml_work_with_s3 => {
+                // TODO: Fix signature calculation mismatch with slashes
+                // const key = "i/am/a/teapot/foo";
+                const key = "foo";
+
+                const bucket = blk: {
+                    const result = try client.call(services.s3.list_buckets.Request{}, options);
+                    defer result.deinit();
+                    const bucket = result.response.buckets.?[result.response.buckets.?.len - 1];
+                    std.log.info("ListBuckets request id: {s}", .{result.response_metadata.request_id});
+                    std.log.info("bucket name: {s}", .{bucket.name.?});
+                    break :blk try allocator.dupe(u8, bucket.name.?);
+                };
+                defer allocator.free(bucket);
+                const location = blk: {
+                    const result = try aws.Request(services.s3.get_bucket_location).call(.{
+                        .bucket = bucket,
+                    }, options);
+                    defer result.deinit();
+                    const location = result.response.location_constraint.?;
+                    std.log.info("GetBucketLocation request id: {s}", .{result.response_metadata.request_id});
+                    std.log.info("location: {s}", .{location});
+                    break :blk try allocator.dupe(u8, location);
+                };
+                defer allocator.free(location);
+                const s3opts = aws.Options{
+                    .region = location,
+                    .client = client,
+                };
+                {
+                    const result = try aws.Request(services.s3.put_object).call(.{
+                        .bucket = bucket,
+                        .key = key,
+                        .content_type = "text/plain",
+                        .body = "bar",
+                        .storage_class = "STANDARD",
+                    }, s3opts);
+                    std.log.info("PutObject Request id: {s}", .{result.response_metadata.request_id});
+                    std.log.info("PutObject etag: {s}", .{result.response.e_tag.?});
+                    defer result.deinit();
+                }
+                {
+                    // Note that boto appears to redirect by default, but java
+                    // does not. We will not
+                    const result = try aws.Request(services.s3.get_object).call(.{
+                        .bucket = bucket,
+                        .key = key,
+                    }, s3opts);
+                    std.log.info("GetObject Request id: {s}", .{result.response_metadata.request_id});
+                    std.log.info("GetObject Body: {s}", .{result.response.body});
+                    std.log.info("GetObject etag: {s}", .{result.response.e_tag.?});
+                    std.log.info("GetObject last modified (seconds since epoch): {d}", .{result.response.last_modified.?});
+                    defer result.deinit();
+                }
+                {
+                    const result = try aws.Request(services.s3.delete_object).call(.{
+                        .bucket = bucket,
+                        .key = key,
+                    }, s3opts);
+                    std.log.info("DeleteObject Request id: {s}", .{result.response_metadata.request_id});
+                    defer result.deinit();
+                }
+                {
+                    const result = try aws.Request(services.s3.list_objects).call(.{
+                        .bucket = bucket,
+                    }, s3opts);
+                    std.log.info("ListObject Request id: {s}", .{result.response_metadata.request_id});
+                    std.log.info("Object count: {d}", .{result.response.contents.?.len});
+                    defer result.deinit();
+                }
+            },
         }
         std.log.info("===== End Test: {s} =====\n", .{@tagName(t)});
     }